Projects

Projects

Tactics Journal

Repo Local Path Purpose
kyleboas/research ~/code/research Pipeline engine — ingest, detect, report, eval, optimization
kyleboas/tacticsjournal.com ~/code/tacticsjournal.com Jekyll site — final publishing surface
kyleboas/tacticsjournal-team ~/code/tacticsjournal-team Orchestration layer — workflows, budgets, team routing
kyleboas/discord-tacticsjournal — Discord bot for Tactics Journal
kyleboas/TacticsAlerts — Tactics alerts
kyleboas/FootballAlerts — Football alerts
kyleboas/TacticsBot — Tactics bot
kyleboas/FPL — Fantasy Premier League

AI Agents & Infrastructure

Repo Local Path Purpose
kyleboas/paperclip ~/code/paperclip Open-source orchestration for zero-human companies
kyleboas/hermes-agent ~/code/hermes-agent The agent that grows with you
kyleboas/hermes-assistant — Hermes assistant
kyleboas/hermes-config — Hermes configuration
kyleboas/secondbrain ~/code/secondbrain A shared memory for AI
kyleboas/ai-dev-tasks — Task management system for AI dev agents
kyleboas/pi-telegram ~/code/pi-telegram Telegram DM bridge extension for pi
kyleboas/pi-docs ~/pi-docs Persistent context docs for AI agents (this repo)

Claw Ecosystem

Repo Local Path Purpose
kyleboas/openclaw — Your own personal AI assistant
kyleboas/nanoclaw — Lightweight alternative to OpenClaw, runs in containers
kyleboas/nullclaw — Fast, small, autonomous AI assistant in Zig
kyleboas/ironclaw — OpenClaw-inspired implementation in Rust
kyleboas/zeroclaw — Claw done right
kyleboas/antfarm — Build your agent team in OpenClaw with one command
kyleboas/nanobot — Ultra-lightweight clawdbot
kyleboas/babyagi3 — BabyAGI

Tools & Utilities

Repo Local Path Purpose
kyleboas/rtk ~/rtk CLI proxy that reduces LLM token consumption by 60-90%
kyleboas/createPRD — PRD generator
kyleboas/if-tools — Infinite Flight tools

Personal Sites & Misc

Repo Local Path Purpose
kyleboas/kyleboas.com — Personal website
kyleboas/images — Image assets
kyleboas/blob — Blob storage
kyleboas/mike-boas — Mike Boas
kyleboas/Erin-Boas — Erin Boas
kyleboas/bertrend — Bertrend
kyleboas/Skinty — Skinty
kyleboas/poll-api — Poll API

Archive / Early Projects

Repo Local Path Purpose
kyleboas/infiniteflight-docs — Documentation for Infinite Flight
kyleboas/discord-py-heroku — Discord bot Heroku starter kit
kyleboas/scriptable — iOS widgets collection
kyleboas/NNWThemesBroadsheet — NetNewsWire theme
kyleboas/IFATC-Education-Group — IFATC Education Group
kyleboas/moltbot-railway-template — Moltbot Railway template
kyleboas/Ralph-Github-Action — Ralph GitHub Action
kyleboas/SyncdChat — Syncd Chat
kyleboas/test — Test repo

Private (no description)

Repo Purpose
kyleboas/docs Docs
kyleboas/Manus Manus
kyleboas/Assistant Assistant
kyleboas/AssistantFiles Assistant Files
kyleboas/Notes Notes

Infrastructure

Tactics Journal Research

Home: https://tacticsjournal.com/research About page (to be moved to /research/about/): ~/code/tacticsjournal.com/research/research.md

  • Railway — Production execution. Logged in as Kyle Boas (railway whoami). Project: research, Environment: production. Services: ingest, detect, rescore, report, autoresearch_hourly, web (dashboard). All services point to the research repo and use PIPELINE_STEP to select which step to run.
  • Railway CLI — Installed (railway v4.35.0). Use railway status, railway logs --service <name>, railway run.
  • Cloudflare AI Gateway — LLM routing for all pipeline calls. Configured via CLOUDFLARE_GATEWAY_URL and CLOUDFLARE_GATEWAY_TOKEN.
  • Cloudflare Dynamic Workers Gateway — Article fetches and YouTube transcript resolution. Configured via DYNAMIC_WORKERS_HOST / DYNAMIC_WORKERS_SECRET. Worker defined in ~/research/cloudflare/dynamic-worker-gateway/.
  • Postgres (pgvector) — Database for sources, embeddings, candidates, reports. On Railway.
  • Paperclip — Agent operating system. API at PAPERCLIP_API_URL. Web UI: https://paperclip-production-067b.up.railway.app/TAC/agents.
  • GitHub — PR-based publishing workflow. Reports land as PRs for Kyle's review. Never auto-merge.
  • Discord — Webhook alerts for new Detect trends and eval runs.

Models (config.json)

  • anthropic/claude-sonnet-4-6 — lead, synthesis, summary, revision
  • workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast — default model, signal, citation, eval
  • openai/text-embedding-3-small — embeddings
  • No OpenRouter or other providers — all routed through Cloudflare AI Gateway

Pipeline Steps (run via make in ~/research)

ingest → backfill → detect → rescore → report
  • make step-ingest — Pull RSS feeds + YouTube transcripts into DB
  • make step-backfill — Fill gaps in source content embeddings
  • make step-detect — Find emerging tactical trends via novelty scoring
  • make step-rescore — Recalculate novelty scores with latest data
  • make step-report — Generate structured research reports from top candidates
  • make dashboard — Start local dashboard at http://localhost:8080/

Autoresearch (Karpathy-style experiment loop)

Canonical loop per stage:

python autoresearch/<stage>/prepare.py   # freeze benchmark
python autoresearch/<stage>/train.py     # edit mutable surface, run, keep improvements

Stages: ingest, detect, report

Production eval/tuning:

  • make eval-report — Offline report-quality evaluation
  • make optimize-ingest-policy — Tune ingest policy (no-LLM heuristics)
  • make benchmark-report — Benchmark report policy on recent reports
  • make optimize-report-policy — Search/apply best report policy
  • make autoresearch-daily — Full no-LLM hourly policy loop

Reference: karpathy/autoresearch and autoresearch-anything (~/code/autoresearch-anything/)

Key Files in ~/research

  • AGENTS.md — Full agent instructions, team roles, budget, morning briefing format
  • README.md — Pipeline documentation, architecture, CLI usage
  • main.py — End-to-end pipeline logic
  • server.py — Dashboard + step runner API
  • detect_detectors.py — Frontier-gap candidate generation
  • detect_policy_config.json — Detect sensitivity and score gates
  • report_policy_config.json — Report quality gates and generation config
  • config.json — Model selection defaults
  • railway.toml — Railway service layout and cron schedules
  • env.example — All environment variables
  • feeds/rss.md — RSS feed list
  • feeds/youtube.md — YouTube channel list