Hermes Memories

Protected compendium generated from Gordon's wiki.

Generated from /opt/data/hermes-pages/wiki

Hermes Memories

A single-page HTML compendium of Gordon's private wiki pages, generated from the existing wiki HTML source. Original page links remain available under /wiki/....

16wiki pages
Protectedclient-side access gate
2026-05-16generated
/wiki/entities/gordon-rouse

Gordon Rouse

Director of Engineering at Kla, based in the Bay Area. Relocating to Ventura, CA (1920 E Linda Vista Ave, 93001) within ~1 year.

Core Facts

  • **Role:** Director of Engineering
  • **Company:** [[KLA Corporation]] — Milpitas, CA
  • **Tenure:** 31 years 5 months (joined ~1994)
  • **Education:** MS Computer Engineering — Santa Clara University; BS Physics — UC Santa Barbara
  • **Location:** Dublin/Santa Clara, CA → Ventura, CA (relocating)
  • **Email:** gordon.rouse@gmail.com
  • **LinkedIn:** linkedin.com/in/gordonrouse
  • **GitHub:** github.com/rousegordon-ops

    • Career Trajectory

    RoleDurationYears
    Director of EngineeringFeb 2026 – Present4 mos
    Engineering Program ManagerJan 2011 – Feb 202615 yrs 5 mos
    Software Engineering Manager2006 – Nov 20115 yrs 11 mos
    Software Engineer1998 – 20068 yrs
    Systems Engineer1995 – 19983 yrs
    Pharmaceutical Sales Rep1993 – 19952 yrs

    Pre-KLA: Spent 2 years as a pharmaceutical sales representative at Syntex Labs before pivoting to engineering. Kla has been the only engineering employer — 31 years of continuous growth within one company.

    Related

  • [[KLA]] — current employer, 31 years
  • [[Ventura relocation]] — relocation and life-planning context
  • [[sidekick-studio]] — hobby project, personal AI team of 8 sidekicks (Mentor, Concierge, CFO, Foodie, Trainer, Stylist, Handyman, Cinephile)
  • [[hobbies/fitness]], [[hobbies/hiking]], [[hobbies/backcountry-fishing]] — active outdoorsman
  • [[hobbies/personal-style]] — takes dressing well seriously, faces a specific fit challenge (big seat, small waist)

    • Personal Finance Snapshot

  • Sold ~$800K of KLA stock at $800/share ~18 months ago. Stock now ~$1,800. Capital used to fund Ventura remodel. Psychologically difficult but financially sound given the cost overruns.
  • Ventura house: ~$2.7M all-in (including $900K remodel), ~$800K remaining mortgage, paid off before 2027.
  • Investable assets: >$3M. Wife earns >$200K/year, continuing to work through retirement.
  • Target retirement: 2027. Financially secure — wife's income covers living expenses while portfolio compounds.
  • See [[ventura-relocation]] for retirement details.
    • /wiki/concepts/ventura-relocation

    Ventura Relocation

    Gordon Rouse is relocating from the Bay Area (Dublin/Santa Clara) to Ventura, CA — specifically 1920 E Linda Vista Ave, Ventura, 93001. Timeline: within ~1 year.

    Why

    The southbound commute math doesn't work from Ventura. All his target employers are South Coast (Goleta, Thousand Oaks, Camarillo), so a northbound commute into the Bay Area would be the wrong direction. If Gordon stays at KLA it would be almost entirely remote — not a driving factor.

    The house renovation is a major undertaking: Ventura Renovation. Gordon designed all the architecture himself. Full remodel including 5 shear walls and 7 underpins.

    Equally important: Sidekick Studio is a hobby project he's building — a personal AI team of 8 sidekicks covering everyday life domains. It's interesting and he's committed to it, but it's not a serious career pivot. A potential long-term asset to have, nothing more.

    Retirement

    Target retirement: 2027. Ventura house fully paid off before then. >$3M investable assets. Wife earns >$200K/year and plans to keep working. This is a financially secure transition, not a leap.

    Target Criteria

  • **Hard constraint:** <30 min commute from Ventura
  • **Employer types:** Semiconductor capital equipment, defense/aerospace, biotech/pharma — places that value his KLA-level systems/firmware/architecture experience
  • **Excludes:** Verilog/VHDL, active clearance, Raytheon/Goleta, adtech, far-remote

    • Target Employers

  • [[Northrop Grumman]] — Camarillo (~26 min)
  • [[Teledyne]] — Thousand Oaks (~37 min)
  • [[Amgen]] — Thousand Oaks (~32 min)

    • Related

  • [[gordon-rouse]] — the person relocating
  • [[KLA]] — current employer, commuting problem
  • [[gordon-rouse-career]] — decision context for the move
    • /wiki/projects/sidekick-studio

    Sidekick Studio

    Gordon's hobby project. A multi-agent AI chat platform where each "sidekick" is a specialized AI assistant with its own personality, system prompt, and optional custom tools.

    Stack

  • **Server:** Express + Socket.io + PostgreSQL
  • **AI:** Claude Sonnet 4.6 (agent loop), Gemini 2.5 Flash (web search with grounding)
  • **Deployment:** Railway
  • **Auth:** JWT (90-day), bcrypt passwords, email verification via Resend

    • The Roster — 8 Sidekicks

    #SidekickIconTaglineDomain
    1Mentor🌟"Growth & wellness"Reflection, career, growth, mental wellness. Loads all other profiles + notes.
    2Concierge🗺️"Places to go and things to do!"Travel, places, events. Has custom tools: search_flights, validate_business, get_directions, search_places. Loads Cinephile and CFO profiles.
    3CFO💰"Dollars and sense"Personal finance, investing, taxes.
    4Foodie🍽️"Gourmet chef & dietitian"Cooking, cuisine, diet planning.
    5Trainer🏋️"Fit happens"Workouts, fitness goals, progress.
    6Stylist🪞"Outfits & personal style"Personal style, wardrobe.
    7Handyman🔧"Home repair & DIY projects"Home improvement, DIY.
    8Cinephile🎬"Movies, shows & books"Film, TV, and book recommendations. Keeps watched list in preferences, never re-recommends. Uses web_search for streaming availability and new releases.

    Key Decisions

  • **Musician removed** — Musician sidekick no longer exists.
  • **Cinephile covers film, TV, and books** — not replaced by Entertainer. The "Shows" app is a `/shows` URL variant of Cinephile with a narrower system prompt focused on TV and film specifically.
  • **CFO (not Finance)** — name is "the buddy that knows about investments and taxes." CFO is correct.
  • **Stylist (not Fashion)** — the app is in `apps/fashion/` but the displayed name is "Stylist". Consistent with Gordon's preference.

    • Design

  • Righteous font for headings and app names (Google Fonts)
  • App names: amber `#c8882a`
  • "SIDEKICK" branding: burnt sienna `#c45a3a`
  • Mobile-first, touch-optimized
  • No tables in responses (mobile readability)
  • Polling, not streaming — client polls every 2.5s
  • Preferences stored as profile blobs per app, injected into system prompt every message

    • Key Differentiator

    Sidekicks search extensively, verify details, and do actual work. "AI that does the work you'd otherwise spend hours doing." Takes time because it's thorough.

    Most Useful So Far

    Concierge — used heavily during vacation planning, which was the impetus for building the product.

    Related

    Fitness — Trainer is the fitness sidekick

    Personal Style — Stylist is the style sidekick

    Source: SidekickStudio repo at github.com/rousegordon-ops/SidekickStudio

    /wiki/projects/alcove-bathtubs

    66" × 36" Alcove Bathtubs

    Standard alcove bathtub dimensions — the most common size for home bathrooms in the US. Fits a 3-wall alcove installation. This page tracks research on available models, specs, and buying considerations.

    Why 66" × 36"?

    The 66" length fits standard 5-foot alcove openings while the 36" width provides more interior space than the narrower 32" alcove models. It's a practical middle ground — spacious enough for soaking, fits standard framing, and widely stocked by major retailers.

    Key Specs to Compare

    • Length × Width (outer): 66" × 36" — confirm rough-in framing
    • Height: typically 19"–21"
    • Water depth: 14"–16" — deeper is better for a true soak
    • Basin length × width: where your body actually sits — often 42"–48" × 24"
    • Water capacity: 45–55 gallons typical
    • Drain side: left-hand vs. right-hand — must match your plumbing
    • Material: acrylic, fiberglass/acrylic composite, or enameled steel
    • Integrated flange: helps waterproof the alcove seam

    Known Models

    Kohler Mariposa — K-1229-L

    Kohler Mariposa K-1229-L alcove bathtub

    image: shopkohlernj.com (authorized Kohler dealer)

    SpecValue
    Dimensions66" L × 35-7/8" W × 21-1/4" H
    Water depth14"
    Basin48" L × 24" W
    DrainLeft-hand
    MaterialAcrylic
    FeaturesIntegral flange, embossed bathing well
    Approx. price$$$ (premium)

    Note: K-1229-L is the flanged variant without an apron. The residential apron version is K-1229-LA-0.

    PROFLO Hillsboro / Plus Series — PFS6636LSKWH

    PROFLO Hillsboro PFS6636LSKWH alcove bathtub

    image: American Bath Group via royalbathplace.com

    SpecValue
    Dimensions66" L × 36" W × 19-3/4" H
    Basin42" L × 24" W
    DrainLeft-hand
    MaterialAcrylic
    FeaturesSkirted front, integrated tile flange
    Approx. price$$ (budget-friendly)

    Note: Ferguson now markets this under their "Plus Series" branding rather than "Hillsboro", but it's the same tub. Full model number includes color suffix (WH = white).

    Aquatic AcrylX — 6636SMIN

    Aquatic AcrylX 6636SMIN alcove bathtub

    image: aquaticbath.com

    SpecValue
    Dimensions66" L × 36" W × 18" H
    MaterialAcrylX (reinforced acrylic)
    FeaturesAbove-the-floor rough, easy install
    DrainRight-hand (or left as ac003093-l-to-wh)
    Water capacity52 gal

    Brand correction: Originally listed as "American Standard AcrylX" but this is actually an Aquatic Bath product (same American Bath Group parent as PROFLO/Ferguson, distinct from American Standard). American Standard's own alcove lineup tops out at 60", not 66".

    Carver Tubs — ALR6636

    Carver Tubs ALR6636 alcove bathtub

    image: carvertubs.com

    SpecValue
    Dimensions66" L × 36" W × 20.5" H
    Water capacity50 gal
    MaterialAcrylic
    DrainRight-hand

    Installation Notes

    • Framing: standard 66" alcove rough-in — verify opening is exactly 66" (not 65.5")
    • Drain alignment: measure from the finished wall to center of drain hole before buying
    • Tile flange: most models have an integral flange on three sides — mount directly against backer board
    • Leveling: adjustable feet or mortar bed — never set directly on subfloor without leveling compound
    • Access panel: required for drain hookup servicing — plan your alcove access

    Material Tradeoffs

    Acrylic: Most common. Lightweight, retains heat well, easy to clean. Reinforced cores (e.g., AcrylX) add stiffness.

    Enameled steel: Cheaper, heavier, cools faster. Chips can expose steel and cause rust.

    Fiberglass/FRP: Budget option. Porous surface can stain; less durable long-term.

    Things to Verify Before Buying

    • Drain side — left-hand vs. right-hand — must match existing plumbing
    • Minimum alcove opening width (some 36" tubs need 36.25" rough-in)
    • Water depth if you prefer a deep soak (14"+ is better than 12")
    • Integrated flange vs. tile flange — determines waterproofing approach
    • Bathewell texture — some prefer smooth, some want slip-resistant texture

    Research Status

    Initial data gathered from Kohler, Ferguson, Aquatic Bath, and Carver Tubs official product pages.

    Still needed:

    • Actual water depth vs. stated — manufacturers sometimes overstate
    • Long-term durability reviews
    • Price tracking across retailers
    • Head-to-head comparison + recommended pick for the project
    /wiki/hobbies/backcountry-fishing

    Backcountry Fishing

    Gordon is an avid backcountry trout fisherman in the Sierra Nevada.

    Preferred Type of Water

    Seeks steep, remote creeks with defined plunge pools — not meadow meanders or flat float water. Wants pocket water and boulder-choked canyons with wild trout and genuine solitude. Hikes in, camps 2–3 nights, hikes out.

    Favorite Waters (Lost / Seeking Alternatives)

    WaterStatus
    **North Fork San Joaquin**Favorite, fire closed the trail
    **South Fork Kings River**Still active
    **Ilillouette Creek**Still active
    **Bubbs Creek** (Kings Canyon)Top pick as replacement
    **Clark Fork Trail**Under consideration

    His benchmark is Sheep Crossing on NFSJ — steep granite canyon, big plunge pools, wild rainbows and browns, total solitude. A fire took out the trail access several years ago and it's now too difficult to reach. He's been searching for a comparable replacement ever since.

    What He Looks For in a Trip

  • **Hike in:** 5–8 miles preferred
  • **Camp:** 2–3 nights, needs flat ground near water, some tree cover, genuine remoteness, and ideally a view
  • **Permits:** Open to lottery locations but prefers easier permit situations
  • **Fishing + camping together** — both must be good; good fishing alone is not enough

    • What He's Done / Explored

  • Bubbs Creek (Kings Canyon): steep bouldery canyon, defined plunge pools, good camp near the creek, ~5–6 miles in
  • Clark Fork: trail starts near river then climbs high on canyon wall, drops back to river at ~5–6 miles for camping
  • Mono Creek (west of Lake Edison, John Muir Wilderness)
  • Carson-Iceberg Wilderness — East Fork Carson River (wild trout section)
  • Wind River Range, Wyoming (South Fork Little Wind River) — grizzly territory
  • Lost Sierra north of Truckee
  • Colorado and New Mexico explored as alternatives — neither matched the Sierra

    • Grizzly Consideration

    The Wind River Range has a growing grizzly population as part of the Greater Yellowstone Ecosystem.

    Related: Hiking, Outdoor Gear

    /wiki/hobbies/backpacking

    Backpacking

    Overnight wilderness trips — the combination of hiking, camping, and backcountry fishing. Gordon goes in deep with his wife, typically 2–3 nights, hiking 5–8 miles in to base camp.

    Active Objective: Tinker Knob

    A Sierra Nevada backpacking route — roughly 14–16 miles with 3,000+ feet of gain — planned for later in the summer 2026. Training hikes with his wife are ongoing.

    Training

  • **Eagle Peak Loop via Mitchell Canyon** — best "pound-for-pound" training on Mount Diablo. ~8 miles, 2,100+ ft gain. Wildflowers in spring.
  • **Mount Diablo Grand Loop** — considered but set aside (too long)
  • Logistics: Mitchell Canyon gate opens 8:00 AM. Full-day outing. Post-hike food at Enye (Latin Table) in Clayton is a recurring favorite.
  • Also interested in: Murietta Falls for vertical training intensity

    • The Full Trip Package

    For Gordon, a good backpacking trip needs both:

    1. Good fishing — steep, remote creeks with plunge pools, wild trout, solitude

    2. Good campsite — flat ground near water, tree cover, genuine remoteness, ideally a view

    Good fishing alone is not enough.

    Key Reference: Sheep Crossing (Lost)

    Benchmark trip: Sheep Crossing on the North Fork San Joaquin. Steep granite canyon, big plunge pools, wild rainbows and browns, total solitude. A fire took out the trail several years ago — access is now too difficult. Has been searching for a comparable replacement ever since.

    What He's Done / Explored

    WaterNotes
    **Bubbs Creek** (Kings Canyon)Top replacement pick. Steep bouldery canyon, plunge pools, good camp. ~5–6 mi in.
    **Clark Fork**Trail starts near river, climbs high on canyon wall, drops back at ~5–6 mi for camping.
    **South Fork San Joaquin**Above Florence Lake — steep canyon, good pocket water
    **Mono Creek** (John Muir Wilderness)West of Lake Edison
    **East Fork Carson River** (Carson-Iceberg)Wild trout section

    Also explored: Wind River Range WY, Lost Sierra north of Truckee, Colorado, New Mexico. None matched the Sierra.

    Related

    Hiking — day hikes (separate from backpacking)

    Backcountry Fishing — fishing component of backpacking trips

    Related: Fitness — conditioning for Tinker Knob

    /wiki/hobbies/hiking

    Day Hikes

    Day hikes — separate from overnight backpacking trips. Gordon hikes with his wife, usually in the Bay Area or nearby regions.

    Recent / Regular

  • **Mount Diablo Eagle Peak Loop via Mitchell Canyon** — ~8 miles, 2,100+ ft gain. Best "pound-for-pound" training hike near the Bay Area. Wildflowers in spring (April window). Mitchell Canyon gate opens 8:00 AM. Post-hike food: Enye (Latin Table) in Clayton.
  • **Morel mushroom hunting** — seasonal, usually spring after rains

    • Gear / Logistics

  • Full-day hike logistics planned for each outing (timing, food, water, driving directions)
  • Typically go as a couple

    • Related

  • [[hobbies/backpacking]] — overnight wilderness trips (Tinker Knob, etc.)
  • [[hobbies/backcountry-fishing]] — backcountry fishing component of backpacking trips
    • /wiki/hobbies/fitness

    Fitness & Health

    Gordon trains consistently, 3–4x per week. His primary goals are maintaining strength and posterior chain flexibility as he approaches 56.

    Training Style

  • **Upper/Lower split** — runs roughly weekly, targets each movement 2x per week
  • Upper body day: bar dips, bent-over row, pull-up, standing press — 2 sets of 5–10 reps each
  • Legs: squats, walking lunges, RDLs — compound focus, auto-regulation (roughly same each week)
  • Sessions run ~30–35 minutes
  • **Auto-regulation:** does roughly the same routine weekly; not rigid about progression but watches for progressive overload drift

    • Health Notes

  • Has a **mild left hip/glute issue** — managed carefully on squats and lunges. No discomfort reported from sessions.
  • Squat progression: warm-up 45 → 95 → 145 (2×6 working sets). When consistently clean, bump to 3 working sets before adding weight.

    • Daily Flexibility Protocol

    Gordon does a short daily mobility routine, broken into two sessions:

    AM — "Wake Up" (movement/lubrication, ~5 min)

  • Elephant Walks (1 min)
  • 90/90 Hip Switches (1 min)
  • Bodyweight Jefferson Curls (1 min)
  • Downward Dog Pedals (2 min) — replaced Bird Dog (back/core stability, not ROM)

    • PM or Post-Workout — "Deepen" (lengthening, ~5 min)

  • Single-Leg Calf Stretch (1 min)
  • Alternating Cossack Squats (2 min)
  • Pigeon Stretch or Figure-4 (2 min)

    • Two 5-minute routines > one 10-minute routine. Frequency over duration for flexibility.

    Also interested in the World's Greatest Stretch (WGS) — good for hip flexors, glutes, hamstrings, and thoracic spine. Thoracic rotation is a bonus for skiing.

    Conditioning Context

    Hiking Tinker Knob (14–16 miles, 3,000+ ft gain) later in summer 2026. Works in with strength training.

    Alcohol Use

    Consciously cutting back. Initially identified a "reward/buffer" cycle — drinking to mark the end of a hard day, then using it as a buffer to wind down before bed.

    Currently doing one drink per night. Feels "a little better" in the morning but:

  • Not sleeping better yet
  • Harder to fall asleep
  • Dreams are more vivid and more likely to wake him up

    • A 9–10 PM work meeting that gets him "spun up" is "really unhelpful" for the wind-down routine.

    Making progress — one drink is a significant win from a nightly buffer habit. Chemical recalibration takes time.

    Related: Hiking, Backpacking

    /wiki/projects/hermes-agent

    Hermes Agent

    Gordon's personal AI agent, built on top of Nous Research's Hermes Agent framework.

    Repo: rousegordon-ops/hermes-agent (public)

    Docs: hermes-agent.nousresearch.com

    Core Features

  • **TUI interface** — full terminal UI with multiline editing, slash-command autocomplete, conversation history, interrupt-and-redirect, streaming tool output
  • **Multi-platform** — Telegram, Discord, Slack, WhatsApp, Signal, CLI; voice memo transcription; cross-platform conversation continuity
  • **Memory & skills** — agent-curated memory, autonomous skill creation after complex tasks, FTS5 session search, Honcho user modeling
  • **Cron scheduling** — daily reports, nightly backups, weekly audits in natural language
  • **Subagents** — spawn isolated subagents for parallel workstreams; Python scripts via RPC
  • **Flexible deployment** — local, Docker, SSH, Daytona, Singularity, Modal backends

    • Gordon's Usage

  • Daily job report cron (Mon-Fri 7 AM PT)
  • Daily weather brief (6 locations)
  • Daily LLM usage stats
  • Todo list management
  • Wiki updates (this wiki!)
  • Job search automation via Venteur API

    • Related

  • [[gordon-rouse]]
  • [[KLA]]
    • /wiki/entities/kla

    KLA Corporation

    Semiconductor capital equipment company — inspection and metrology tools for wafer fabs. Headquartered in Milpitas, CA. NYSE: KLAC.

    Gordon's Tenure

    Gordon Rouse has been at KLA for 31 years 5 months — the only engineering employer of his career. Joined as Systems Engineer (1995), moved through Software Engineer → Software Engineering Manager → Engineering Program Manager → Director of Engineering.

    The progression shows strong internal mobility and sustained technical growth within one company — unusual in today's environment.

    Context

    KLA's commute from Gordon's current Dublin/Santa Clara location is 40–60 min each way. Post-Ventura relocation, this becomes 80+ min (wrong direction on the train). The math breaks, which is the primary driver of Gordon Rouse's exploration.

    Related

  • [[gordon-rouse]] — 31-year employee, currently Director of Engineering
  • [[Ventura relocation]] — the relocation that makes the KLA commute untenable
    • /wiki/raw/articles/karpathy-llm-wiki

    LLM Wiki

    A pattern for building personal knowledge bases using LLMs.

    This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

    The core idea

    Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

    The idea here is different. Instead of just retrieving from raw documents at query time, the LLM incrementally builds and maintains a persistent wiki — a structured, interlinked collection of markdown files that sits between you and the raw sources. When you add a new source, the LLM doesn't just index it for later retrieval. It reads it, extracts the key information, and integrates it into the existing wiki — updating entity pages, revising topic summaries, noting where new data contradicts old claims, strengthening or challenging the evolving synthesis. The knowledge is compiled once and then kept current, not re-derived on every query.

    This is the key difference: the wiki is a persistent, compounding artifact. The cross-references are already there. The contradictions have already been flagged. The synthesis already reflects everything you've read. The wiki keeps getting richer with every source you add and every question you ask.

    You never (or rarely) write the wiki yourself — the LLM writes and maintains all of it. You're in charge of sourcing, exploration, and asking the right questions. The LLM does all the grunt work — the summarizing, cross-referencing, filing, and bookkeeping that makes a knowledge base actually useful over time. In practice, I have the LLM agent open on one side and Obsidian open on the other. The LLM makes edits based on our conversation, and I browse the results in real time — following links, checking the graph view, reading the updated pages. Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase.

    This can apply to a lot of different contexts. A few examples:

  • **Personal**: tracking your own goals, health, psychology, self-improvement — filing journal entries, articles, podcast notes, and building up a structured picture of yourself over time.
  • **Research**: going deep on a topic over weeks or months — reading papers, articles, reports, and incrementally building a comprehensive wiki with an evolving thesis.
  • **Reading a book**: filing each chapter as you go, building out pages for characters, themes, plot threads, and how they connect. By the end you have a rich companion wiki. Think of fan wikis like Tolkien Gateway — thousands of interlinked pages covering characters, places, events, languages, built by a community of volunteers over years. You could build something like that personally as you read, with the LLM doing all the cross-referencing and maintenance.
  • **Business/team**: an internal wiki maintained by LLMs, fed by Slack threads, meeting transcripts, project documents, customer calls. Possibly with humans in the loop reviewing updates. The wiki stays current because the LLM does the maintenance that no one on the team wants to do.
  • **Competitive analysis, due diligence, trip planning, course notes, hobby deep-dives** — anything where you're accumulating knowledge over time and want it organized rather than scattered.

    • Architecture

    There are three layers:

    Raw sources — your curated collection of source documents. Articles, papers, images, data files. These are immutable — the LLM reads from them but never modifies them. This is your source of truth.

    The wiki — a directory of LLM-generated markdown files. Summaries, entity pages, concept pages, comparisons, an overview, a synthesis. The LLM owns this layer entirely. It creates pages, updates them when new sources arrive, maintains cross-references, and keeps everything consistent. You read it; the LLM writes it.

    The schema — a document (e.g. CLAUDE.md for Claude Code or AGENTS.md for Codex) that tells the LLM how the wiki is structured, what the conventions are, and what workflows to follow when ingesting sources, answering questions, or maintaining the wiki. This is the key configuration file — it's what makes the LLM a disciplined wiki maintainer rather than a generic chatbot. You and the LLM co-evolve this over time as you figure out what works for your domain.

    Operations

    Ingest. You drop a new source into the raw collection and tell the LLM to ingest it. The LLM reads the new source, extracts key information, identifies which existing pages need to be updated or created, makes the changes, updates the index, and logs the action. Very little of this is mechanical — the LLM has to understand the source, compare it against the current state of the wiki, and make good decisions about what to create, update, or leave alone. This is where the LLM's intelligence really matters. A bad ingest creates noise; a good ingest strengthens the whole structure.

    Query. You ask the wiki a question. The LLM reads the index, identifies relevant pages, reads them, synthesizes an answer, and responds. In many cases the answer is already fully in the wiki — the LLM just reads and summarizes. In some cases the question touches things that haven't been synthesized yet, so the LLM does some creative work to connect the dots. In rare cases the question reveals a gap in the wiki that warrants a new page. When that happens, the LLM creates the page and updates the index, so future questions can retrieve that synthesis directly.

    Maintain. The LLM periodically audits the wiki: finds broken links, orphaned pages, stale content, contradictions, and fixes them. This is like a periodic garbage collection pass. The wiki accumulates debt over time without this — links go stale, entities drift, and the index becomes incomplete. The maintenance pass keeps the wiki healthy.

    Schema design

    The schema defines how the wiki is organized, what conventions to follow, and what workflows to use. It evolves with the wiki — as you discover new patterns and conventions, you update the schema so the LLM applies them consistently going forward.

    A typical schema defines:

  • **The domain** — what the wiki covers. The scope doesn't have to be fixed forever, but it anchors everything.
  • **File naming conventions** — e.g. lowercase-hyphens, no spaces, date-prefixed for logs.
  • **Required frontmatter** — what fields every wiki page must have (title, tags, sources, dates, etc.).
  • **Layer conventions** — raw/ is immutable; wiki pages are owned by the LLM; schema defines the rules.
  • **Tag taxonomy** — a controlled vocabulary of tags. Freeform tags are noise. The taxonomy keeps the wiki scannable.
  • **When to create pages** — e.g. only when an entity is mentioned in 2+ sources, or is central to one source.
  • **Cross-referencing rules** — e.g. every page must link to at least 2 other pages.
  • **Provenance conventions** — how to mark which claims come from which sources.
  • **Update policy** — how to handle contradictions, stale content, and page decay.
  • **Thresholds** — when to split, archive, or merge pages.

    • Workflows

    The schema should define workflows for the core operations:

    Ingest workflow: 1) Capture raw source. 2) Discuss takeaways. 3) Check what already exists. 4) Write or update wiki pages. 5) Update index and log. 6) Report changes.

    Query workflow: 1) Read index. 2) Read relevant pages. 3) Synthesize answer. 4) File valuable answers back as pages. 5) Update log.

    Maintain workflow: 1) Find broken links and orphans. 2) Check index completeness. 3) Validate frontmatter. 4) Detect contradictions. 5) Flag stale content. 6) Check source drift. 7) Report findings.

    Why this works

    The wiki becomes more useful over time. Early on it's sparse, but every source you add and every question you ask makes it denser. Cross-references accumulate. Contradictions get surfaced. Synthesis deepens. You stop paying the cost of re-deriving knowledge on every query — instead you pay a one-time ingest cost and then get to use the compiled result forever.

    The LLM does the bookkeeping so you don't have to. You explore and direct; the LLM organizes and maintains. This division of labor is what makes it scale.

    Obsidian

    This pattern pairs naturally with Obsidian. The wiki is just a directory of markdown files, so you can open it as an Obsidian vault and get:

  • Clickable [[wikilinks]] that render as internal links
  • Graph view that visualizes the knowledge network
  • Search across all pages
  • Dataview plugin for querying pages by frontmatter (tags, type, date, etc.)
  • Daily notes plugin for journal-style entries
  • Templates for consistent page creation

    • The LLM does all the writing; you use Obsidian to browse, search, and explore. Your Hermes Agent session writes the wiki; you read it in Obsidian.

    Starting out

    The first thing to do is just start. Drop in a few sources, tell the LLM to ingest them, and see what pages it creates. Don't overthink the schema at first — start simple, let conventions emerge from what the LLM does naturally. Update the schema later based on what works. The wiki will feel sparse at first; stick with it. The density compounds. By the time you've added 20+ sources and asked 50+ questions, you'll have something genuinely useful — a compiled picture of a topic area that would have taken months to build manually.

    The second thing to know: the wiki requires maintenance. Schedule periodic audit passes — fix broken links, check for contradictions, update stale content, keep the index complete. Without maintenance the wiki decays. With maintenance it stays healthy and useful.

    The third thing: be selective about what goes in. The wiki captures synthesis, not noise. A passing mention in a source probably doesn't warrant a page. An entity mentioned across 5 sources and central to your interests absolutely does. The threshold for creating a page should be high enough that every page feels worth having, but low enough that the wiki actually captures everything important.

    /wiki/hobbies/london-trip

    London Trip (July 5–8, 2026)

    Gordon and his wife visit London for 3 nights — arrive morning of July 5, depart morning of July 8.

    Preferences

  • Walking tours and urban exploration
  • Theater (West End)
  • Museums and history
  • Cool public spaces
  • Nice dinners
  • Wife is "crazy for Indian food"

    • Itinerary Highlights

  • **Day 1:** West End matinee → Jamavar (Michelin-star Indian, Mayfair) dinner
  • **Day 2:** Thames Path → Borough Market → Leadenhall Market → Dishoom dinner (Borough Market)
  • **Day 3:** Considered but not yet finalized — options included Spitalfields, Columbia Road Flower Market,optional West End evening show

    • Shows Seen / Booked

  • Hadestown (Sunday July 5 matinee) — booked

    • Hotel

    Staying at Wilde London Paddington (4 N Wharf Rd, W2 1NW). Covent Garden location, ~10 min walk from family in Little Venice, good early check-in prospects.

    Lessons Learned

  • Assistant initially mislabeled Day 1 as "Sunday, July 6" — Gordon caught it immediately and corrected. Note: July 5 is actually a Sunday in 2026.
  • Tube commute from Little Venice to Covent Garden is ~25–30 min, one change. Easy.
  • Europa House Little Venice was considered but Wilde Covent Garden preferred.

    • Related: Travel, Entertainment

    /wiki/hobbies/personal-style

    Personal Style

    Gordon dresses well but faces a specific and common fit challenge: big seat, smaller waist. This drives most of his clothing decisions.

    The Fit Problem

    Standard pants are cut for a flatter seat. When you've got glutes that fill out the seat, you're constantly choosing between:

  • Waist fits → too tight in the seat
  • Seat fits → too big in the waist (the "Size Up Syndrome")

    • The result is a "front dip" — the waistband angles down toward the fly because glutes pull the back up, dragging fabric from the front.

    Also wears pants low — they rest where his body naturally widens. Can't change that height.

    What's Worked

  • **Levi's 541 Athletic Taper** — benchmark fit. Extra room through seat and thigh, tapered below the knee. Most standard brand that handles the geometry.
  • **Barbell Apparel Straight Athletic Fit (34" waist)** — nuclear option. Founded specifically for the big-seat/small-waist build. Technical and smooth fabric. Wife approved.
  • **Barbell Anything Chino** — same fit advantage, chino version.
  • **7 For All Mankind** — stretch denim, athletic builds
  • **AG Jeans Tellis / Graduate** — good seat room without excess thigh

    • What's Been Tried

  • Bonobos Athletic Fit — no brick-and-mortar near him to try; passed on jeans due to online-only friction
  • Fresh Clean Threads tees — Large Tall ordered, "no panacea" (neckline too high/stiff, looked like an undershirt). XL Tall as backup.
  • Lululemon and Vuori — didn't solve the front-dip problem

    • Clothing Preferences

  • **Henleys over plain tees** — more "intentional shirt" look, adds ruggedness. Olive green + gray hair is a "powerhouse combination."
  • **White sneakers** — clean, modern, works with dark denim
  • **Stone/taupe pants** — pairs well with olive tops and white sneakers
  • **Crew neck preferred** — clean, unfussy, age-appropriate

    • Brands to Try

  • Levi outlet (Livermore Premium Outlets) — 541
  • 7 For All Mankind, True Religion, Diesel — Livermore
  • Polo Ralph Lauren, Calvin Klein, Banana Republic Factory — tees at Livermore outlets

    • Related: Fitness

    /wiki/projects/ventura-renovation

    Ventura House Renovation

    Status: In progress — targeting move before end of 2026

    A major whole-house remodel of the future Ventura home. Gordon did all the architectural design himself.

    Scope:

  • Complete interior remodel
  • 5 shear walls
  • 7 underpins

    • Plan after move: Exterior work remains, including the back deck, side yard stairs, back yard hardscape, and other items. The interior must be done before they can move in.

    Architectural plan: View on Google Drive

    /wiki/log

    Wiki Log

    > Chronological record of all wiki actions. Append-only.

    [2026-05-09] create | Wiki initialized

  • Domain: Personal knowledge base for Gordon Rouse
  • Structure created with SCHEMA.md, index.md, log.md
  • Ingested: karpathy-llm-wiki.md (source pattern reference)
  • Created: [[gordon-rouse]]

    • [2026-05-09] ingest | Karpathy LLM Wiki pattern

  • Files created: raw/articles/karpathy-llm-wiki.md
  • Pages created: SCHEMA.md, index.md, log.md
  • This wiki follows the Karpathy LLM Wiki pattern — layered architecture with raw/ immutable sources, wiki pages owned by the LLM, and a schema defining conventions.

    • [2026-05-10] create | Acquire a boring local service business deep dive

  • Created: business-opportunities/acquire-local-service-business
  • Updated: business-opportunities, index
  • Deep dive covers Ventura County demand signals, target niches, acquisition filters, diligence questions, and search strategy.

    • [2026-05-10] update | Corrected business opportunity hierarchy

  • Removed Acquire a Boring Local Service Business from the top-level wiki index.
  • Kept it as a child page linked from the relevant idea bucket item on Other Career and Business Opportunities.

    • [2026-05-10] create | Manufacturing / semicap workflow automation deep dive

  • Created: business-opportunities/manufacturing-semicap-workflow-automation
  • Updated: business-opportunities
  • Deep dive covers semicap workflow wedges, productized offers, implementation pattern, target customers, risks, and validation plan.

    • [2026-05-10] update | Rename career-opportunities page label

  • Renamed visible page title/link from “Career Opportunities” to “KLA-Adjacent Local Roles” to clarify it covers Ventura-area employment roles similar to Gordon’s KLA experience.
  • Kept URL at /wiki/career-opportunities so the weekday job report can continue updating it.

    • [2026-05-10] update | Rename broader opportunities page

  • Renamed visible page label from “Other Career and Business Opportunities” to “Other Career and Business Opportunities” to distinguish it from KLA-adjacent local employment roles.
  • Kept URL at /wiki/business-opportunities.

    • [2026-05-10] update | AI consulting market trend scan

  • Updated Other Career and Business Opportunities with a dated trend scan: agentic workflow automation, production RAG quality, domain-specific systems, pricing/packaging shifts, and productized consulting pilots.
  • Updated Manufacturing / Semicap Workflow Automation with market trend confirmation, auditable agentic workflow pilot positioning, proof-of-work expectations, and evaluation-harness guidance.

    • [2026-05-10] update | Improve wiki text contrast

  • Replaced low-contrast muted gray text styles across wiki pages with higher-contrast readable text colors.
  • Updated the reusable wiki maintenance skill to avoid low-contrast formatting for substantive content.

    • [2026-05-10] create | AI workflow automation for local SMBs deep dive

  • Created: business-opportunities/ai-workflow-automation-local-smbs
  • Updated: Other Career and Business Opportunities
  • Deep dive covers local SMB target types, concrete workflow examples, productized offers, screening criteria, and go-to-market plan.

    • [2026-05-10] refactor | Other Career and Business Opportunities topology

  • Reworked Other Career and Business Opportunities into a concise portfolio hub rather than a duplicated analysis page.
  • Created parent child hub: AI Consulting & Workflow Automation for shared AI consulting thesis, market signals, packaging, geography, and validation plan.
  • Moved vertical deep dives under that hub: Manufacturing / Semicap Workflow Automation and AI Workflow Automation for Local SMBs.
  • Left old vertical URLs as redirects to preserve existing links.

    • [2026-05-10] update | Light industrial implementation stack

  • Updated AI Workflow Automation for Local SMBs with detailed implementation guidance for the light industrial / manufacturing / distribution use case.
  • Added recommended tech stack, integration moat, per-use-case pipelines, eval/observability requirements, and explicit V1 exclusions.

    • [2026-05-10] update | Linked third-party implementation tools

  • Updated AI Workflow Automation for Local SMBs so third-party tools in the light industrial / manufacturing / distribution implementation example link to official docs, company descriptions, or relevant project pages.

    • [2026-05-10] refactor | SMB implementation drilldowns

  • Moved light industrial / manufacturing / distribution implementation details out of the parent SMB page into a dedicated drilldown linked from the first column of the target-types table.
  • Created a senior care / home health admin / care coordination implementation drilldown with recommended stack, integrations, use-case pipelines, guardrails, and linked third-party tools.

    • [2026-05-10] create | Remaining SMB implementation drilldowns

  • Created implementation drilldowns linked from the SMB target-types table for home-service contractors, property management / real estate services, construction / renovation / specialty trades, medical/dental/veterinary/therapy practices, professional services, and compliance-heavy local services.
  • Each drilldown includes stack guidance, linked third-party tools/vendors, integration targets, concrete workflow pipelines, and V1 guardrails.

    • [2026-05-10] create | Manufacturing / semicap implementation drilldowns

  • Created implementation drilldowns linked from the Manufacturing / Semicap Workflow Automation opportunity table for field issue triage + service knowledge capture, ECO/ECR copilot, configuration / release readiness monitor, and program status automation.
  • Each drilldown includes stack guidance, linked third-party tools/vendors, integration targets, workflow pipelines, and V1 guardrails.
    • /wiki/schema

    Wiki Schema

    Domain

    Personal knowledge base for Gordon Rouse — career, technology, life decisions, and personal development.

    Conventions

  • File names: lowercase, hyphens, no spaces (e.g., `gordon-rouse.md`)
  • Every wiki page starts with YAML frontmatter (see below)
  • Use `[[wikilinks]]` to link between pages (minimum 2 outbound links per page)
  • When updating a page, always bump the `updated` date
  • Every new page must be added to `index.md` under the correct section
  • Every action must be appended to `log.md`
  • **Provenance markers:** On pages that synthesize 3+ sources, append `^[raw/articles/source-file.md]`

    • at the end of paragraphs whose claims come from a specific source.

    Frontmatter

    
---
title: Page Title
created: YYYY-MM-DD
updated: YYYY-MM-DD
type: entity | concept | comparison | query | summary | meta
tags: [from taxonomy below]
sources: [raw/articles/source-name.md]
confidence: high | medium | low
contested: true
contradictions: [other-page-slug]
---

    Tag Taxonomy

  • `entity`: People (GORDON, KLA, employers)
  • `concept`: Technical topics, career concepts
  • `life`: Career decisions, relocation, life choices
  • `meta`: Schema, index, log entries

    • Page Thresholds

  • **Create a page** when an entity/concept appears in 2+ sources OR is central to one source
  • **Add to existing page** when a source mentions something already covered
  • **DON'T create a page** for passing mentions, minor details, or things outside the domain

    • Wiki Gardener Policy

    The agent should infer taxonomy and hierarchy from the content instead of requiring Gordon to specify it.

  • Parent pages are maps, not dumping grounds. Keep hub/index pages concise and navigable.
  • Promote repeated or central topics to pages. If a topic appears repeatedly, becomes a decision area, or is requested as a deep dive, create a dedicated page.
  • Split by semantic containment. Broad domain → hub page; major bucket → child hub; specific opportunity/idea/company/source → child page.
  • Split pages that exceed ~300–500 words in a section or contain multiple independent subtopics.
  • Create child pages for deep dives by default. Example: Other Career and Business Opportunities → AI Consulting → Engineering Knowledge Base + RAG.
  • Preserve navigation. When creating/splitting pages, add links from parent to child and child back to parent, plus related pages.
  • Refactor without asking when the structure is obvious. If uncertain, make a reasonable structure and note it can be adjusted later.
  • Do not regenerate the wiki from markdown. Gordon’s current wiki is maintained as static HTML under /opt/data/hermes-pages/wiki and deployed directly to Cloudflare Pages.

    • Entity Pages

    One page per notable entity. Include:

  • Overview / what it is
  • Key facts and dates
  • Relationships to other entities ([[wikilinks]])
  • Source references

    • Update Policy

    When new information conflicts with existing content:

    1. Check the dates — newer sources generally supersede older ones

    2. If genuinely contradictory, note both positions with dates and sources

    3. Mark the contradiction in frontmatter: contradictions: [page-name]