README.md

Travis Gilbert v26, Flint, MI

Writer, researcher, self-taught developer. I build tools that think about information and I make videos about whatever makes me curious.

build: passingsubscribers: 30kstatus: building in publicvideos: 70+

I don't have a computer science degree. I do have projects.

United States Patent Application, 2026/0322

Abstract

A method and apparatus for investigating whatever is interesting, comprising: a curiosity-driven video production pipeline (70+ videos, 30,000 subscribers), an epistemic intelligence engine capable of measuring its own cognition across seven axes, a knowledge management workbench designed for a brain that does not hold still, a self-improving plugin ecosystem with Bayesian confidence updating across 250+ typed knowledge claims, a government property sales portal, a compliance tracking system that consolidated a seven-step six-software SOP into four button presses, and a community music festival serving 3,000 attendees with 30+ musical acts and 50+ vendors annually.

The system operates under persistent resource constraints (see: Known Limitations) and compensates through the construction of external cognitive instruments including production pipelines, task management architectures, writing workflows, and measurement frameworks.

The inventor is based in Flint, Michigan. The inventor does not have a computer science degree. The inventor does have projects.

Theseus EngineIQ 17.6 → 35.7
PIPELINE7 passes
1# SBERT embedding
2# BM25 lexical
3# NLI classification
[REDACTED]Level 4–8 capabilities
[REDACTED]# structural similarity
5# spaCy NER
6# community detection
7# IQ measurement
[REDACTED]# GNN link prediction
[REDACTED]# counterfactual sim
OUTPUT
knowledge_graph# neurons + weights
COMPUTE
ONNX Runtime# frequent inference
Modal GPU# training jobs
[REDACTED]# LoRA fine-tuning
Objects = Neurons, Edges = Weights

Description

If there's anything I've learned about attention spans it's that most of it can be chalked up to "X should behave like Y, BUT..."

We are all incredibly interested in contradictions.

A world that makes sense to us entirely is boring. If you sit down and watch a movie and you think of it as one logical cohesive string of events, you didn't really watch a movie. You watched the equivalent of someone's grocery list. The difference between a grocery list and a story is someone saying "I needed to get bananas, BUT..."

Some of you reading this will think "well Travis, what about confirmation bias and echo chambers?" To which I will say: it seems like you've got the start of a good story.

The second important thing I've learned is that multiple things can be true at once. The world resists simplicity no matter how much I would like simple explanations. I was once told that if you can't explain a thing simply, you simply don't understand it.

This is complete nonsense.

An explanation is a model of the world, and as long as we live in a world that is complex, dynamic, ever changing, ever moving towards entropy, the world will resist simplicity. And in exchange for that simplicity (note: exchange is an oversimplification) you actually get nothing in return. What you get is a chance at understanding something much richer than simplicity, which is the world around you.

Index-APIknowledge backend
APPSDjango + DRF
commonplace/# core models
connections/# edge manager
EPISTEMIC
claims# extracted assertions
epistemic_models# belief structures
questions# open threads
tensions# stored contradictions
INFRA
pgvector# semantic search
PostGIS# geographic queries
Redis / RQ# task queues
Railway, PostgreSQL

How I Think

Mostly, in terms of constraints. It's a curse but occasionally fruitful.

When life hands me a set of problems to solve, my immediate thought process is something like: "Given my resources (historically, unimpressive) (Constraint #001), how can I reduce and mitigate damages and buy time?"

I also build tools that compensate for my own cognitive limitations. I have ADHD, OCD, and dyslexia. My working memory is unreliable. My ability to hold complex systems in my head is, for some reason, not. So I build external infrastructure: production pipelines, task management architectures, writing workflows, measurement frameworks.

These aren't workarounds. They're cognitive instruments, and designing them is a discipline I take seriously.

CommonPlaceknowledge workbench
FRONTENDNext.js + React 19
Library/# cluster cards, D3 graphs
ObjectRenderer/# polymorphic by type
Compose/# authoring + engine terminal
Sidebar/# Cmd+K, Resurface
APIIndex-API (Django + DRF)
objects/# notes, sources, hunches
connections/# Theseus-generated edges
epistemic/# models, claims, tensions
DESIGN
rough.js# hand-drawn elements
D3 force# cluster visualization
DotGrid# spring-physics canvas
Tailwind v4, Vercel + Railway
Claims

What is Claimed is:

  1. 1.

    Theseus: Epistemic Intelligence Engine

    An epistemic intelligence engine comprising a seven-pass pipeline that discovers semantic, structural, and logical connections across heterogeneous knowledge objects, with self-measurement across seven axes. IQ: 17.6 to 35.7 in a single activation session. Named for Claude Shannon's maze-navigating mouse.

    spaCySBERTBM25NLIKGE/RotatEONNXPyTorchModal
    Theseus
    PIPELINE
    1..7SBERT → BM25 → NLI → KGE → NER → CD → IQ
    OUTPUT
    knowledge_graph
    ONNX + Modal
  2. 2.

    Index-API: Knowledge Backend

    A Django REST Framework API serving 22 endpoints with 190 tests, comprising object management, epistemic model CRUD, claim extraction, tension tracking, and spaCy-powered connection engine. API-key gated with middleware authentication.

    DjangoDRFspaCypgvectorPostGISRedisRQ
    Index-API
    APPS
    notebook/12 models
    api/22 endpoints
    INFRA
    PostgreSQLRailway
    190 tests
  3. 3.

    CommonPlace: Knowledge Management Interface

    A split-pane knowledge workbench with recursive binary tree layout, polymorphic object rendering (10 types), D3 force-directed cluster visualization, and real-time API integration with the Index-API backend.

    Next.jsReact 19D3rough.jsTailwind v4
    CommonPlace
    FRONTEND
    Library/clusters + D3
    Compose/authoring
    API
    objects/10 types
    Vercel + Railway
  4. 4.

    Codex Plugins: Epistemic ML Development Tools

    A self-improving plugin ecosystem for Claude Code with Bayesian confidence updating across 250+ typed knowledge claims, session-level observation logging, and 11 specialist plugins spanning ML, NLP, graph theory, and UI design.

    Claude CodeBayesianJSONLMarkdownYAML
    Codex
    SURFACES
    chat + code2 surfaces
    PLUGINS
    11 specialists250+ claims
    Self-improving
  5. 5.

    Publishing API: Writing Studio Backend

    A Django content management system with HTMX editor, markdown toolbar, visual pipeline (Draft to Published), and GitHub Contents API integration for automated deployment. Includes video production pipeline with 7 batch phases.

    DjangoHTMXPillowGitHub APITailwind
    Studio
    DJANGO
    content/pipeline
    EDITOR
    HTMXsplit-pane
    GitHub API deploy
  6. 6.

    GitHub-MCP: Custom MCP Server

    A Model Context Protocol server bridging GitHub API operations (issues, PRs, commits, releases) into Claude Code tooling. Handles authentication, pagination, and rate limiting transparently.

    TypeScriptMCP SDKGitHub APIOAuth
    GitHub-MCP
    MCP
    tools/CRUD ops
    authOAuth + rate limit
    Model Context Protocol
  7. 7.
    GCLBA Property Sales Portal

    A public-facing property sales portal for the Genesee County Land Bank Authority. Dynamic forms per program type, application scoring, document upload and tracking, reCAPTCHA v3 integration.

    DjangoHTMXPostgreSQLRailway
    GCLBA Portal
    DJANGO
    programs/4 types
    forms/dynamic
    apply.thelandbank.org
  8. 8.
    Compliance.Thelandbank.org

    A compliance tracking system that consolidated a seven-step, six-software SOP into four button presses. Manages 2,000+ property records with automated workflow, FileMaker integration, and PostgreSQL backend.

    DjangoFileMakerPostgreSQL
    Compliance
    DJANGO
    tracking/2K+ homes
    filemaker/bridge
    7 steps to 4 buttons
  9. 9.

    Compliance Inspection Tracker

    A mobile-first inspection tracking tool for field compliance officers. Offline-capable photo capture, GPS tagging, and sync-on-reconnect architecture for areas with unreliable connectivity.

    DjangoPWAIndexedDBService Worker
    Inspector
    PWA
    offline/sync-on-reconnect
    capture/GPS + photo
    Mobile-first
  10. 10.
    travisgilbert.me

    This website. A Next.js 16 static site with App Router, rough.js hand-drawn elements, deterministic PRNG generative art, five content collections with Zod validation, and seven custom fonts. Server Components by default, client only when needed.

    Next.js 16React 19rough.jsTailwind v4Zod
    travisgilbert.me
    FRONTEND
    app/5 collections
    rough.jshand-drawn
    Vercel SSG
  11. 11.
    Curious Tangents: YouTube Channel

    30,000 subscribers, 70+ videos investigating how design decisions shape human outcomes. Covers topics from urban planning to information architecture. Full production pipeline: research, scripting, recording, editing, publishing.

    DaVinci ResolveUlyssesTickTickYouTube API
    Curious Tangents
    PIPELINE
    P0..P7research to publish
    70+ videos30K subs
    Investigation-driven
  12. 12.

    Porchfest: Community Music Festival

    A community music festival serving 3,000 attendees annually with 30+ musical acts and 50+ vendors. Coordination of logistics, vendor management, scheduling, volunteer organization, and public communications.

    Event PlanningLogisticsCommunity
    Porchfest
    EVENT
    30+ acts50+ vendors
    3,000attendees
    Annual, Flint MI
Compliance Systemcompliance.thelandbank.org
DJANGO
compliance/# tracking engine
models.py# 2K+ homes
views.py# SOP automation
INTEGRATION
filemaker/# fixed connection
postgresql
7 steps to 4 buttons
Prior Art
Claude ShannonInformation theory, maze-solving mouse (Theseus), the idea that information can be measured.
Vannevar Bush"As We May Think" (1945). The memex. Associative trails. The original knowledge workbench.
Edward TufteData-ink ratio. Chartjunk. The conviction that visual evidence should be dense, honest, and beautiful.
Jane JacobsEyes on the street. Systems that self-organize from local knowledge. Cities as complex adaptive systems.
Codex Pluginsepistemic ML
TWO-SURFACE ARCHITECTURE
chat_skills/# planning surface
claude_code/# implementation surface
EPISTEMIC LAYER
knowledge/# claims.jsonl + tensions
confidence# Bayesian updating
session_log/# per-session observations
PLUGINS11 specialists
ml-pro# PyTorch + GNNs
scipy-pro# NLP + graph theory
ui-design-pro# 140 claims @ 0.667 conf
django-engine-pro# 111 claims
d3-pro# Observable canon
animation-pro# spring physics
Self-improving development tools

Known Limitations

ADHDWorking memory is unreliable. Compensated through external cognitive instruments (see: every project above).
OCDPattern recognition is weaponized against its owner. Occasionally useful for quality control.
DyslexiaText processing is adversarial. Compensated through careful tooling and workflow design.
No CS DegreeSelf-taught through building. Every project above is evidence of learning by doing.
ResourcesHistorically unimpressive. Compensated through constraint-driven design and creative problem solving.

What I'm Looking For

Mentally stimulating work. Reasonable people. Fair compensation.

I want to work on problems where the tools don't exist yet, where the answer isn't obvious, and where building the right instrument for the job is part of the job. I'm looking for teams that value clarity over ceremony, that understand the difference between rigor and rigidity, and that let people do their best thinking.

If you have something that fits, I'd like to hear about it.

Installation

# contact
email travisgilbert.me@gmail.com
youtube @curioustangents
github @travisgilbert

# prefer email. I read everything.