TradingMapClaw is a dual-engine, multi-model AI research system for US equities and crypto — built by one operator, on one Mac mini, for ~$55 a month.
Every night from dusk to dawn it runs 115 scheduled jobs — pre-market deep dives, a hidden-gem alpha radar, a five-layer supply-chain peel, and a three-model war room that votes on the contested calls. Then it hands you the verdict. Scroll down to watch a night unfold.
One model can be confidently wrong. Two engines catch it. A council of three decides.
Not pity. Visibility.
The Story
I built a production-grade AI research system with no prior coding background, in two months, on a single Mac mini — a dual-engine pipeline that runs 115 scheduled jobs a night and cross-examines its own conclusions before I ever read them.
That is the headline. The background only matters because of the contrast.
Before I was twenty, a hit-and-run collision tore the brachial plexus in my right arm — the bundle of nerves running from the spine into the shoulder. It doesn't heal when it's torn. My right hand lost roughly 95% of its function. Three surgeries. Then I relearned everything with my left hand: shoelaces, writing, typing. I sat the national college entrance exam one-handed. I got in.
Inside the back offices of Wells Fargo, Deutsche Bank, UBS, JPMorgan, and eToro, I did the work with one hand and asked for no lowered standard.
Then ulcerative colitis took my colon. Five more surgeries across eight years — eight operations in total. Today I live with a permanent ostomy for the rest of my life. One body, rebuilt more than once.
In May 2026, I started building. AI became my engineering partner — not a crutch, a collaborator. I described what I wanted; it wrote the first draft; I tested, broke, and hardened it. Two months later: TradingMapClaw. It runs at 4 AM while I sleep, and it does not know or care that it was built by a man with one hand. That is exactly the point.
People living with brachial plexus injury, ulcerative colitis, IBD, Crohn's, and permanent ostomies are nearly invisible in the working world — assumed to be fragile, or gone. I didn't disappear. If you're living with any of this: you are not alone. Not anymore.
Who This Is For
If you recognize yourself here, you already understand why a $55/month research engine matters.
You spent a career reading filings and models by hand. The analysts around you are quietly shipping AI workflows. This is what the other side of that fear looks like — built, running, auditable.
The gap isn't intelligence. It's tooling. Close it.
Not a toy demo — a multi-engine pipeline with fallback chains, quality gates, budget guardrails, and cross-verification, documented from a system that actually runs in production.
Stop starting from a blank file. Start from a blueprint.
A transparent pipeline that gathers from 12+ sources, scores candidates, and reports — for the price of a couple of lunches a month, not thousands per seat.
Institutional-style depth. Solo-operator budget.
This system is proof that constraint is not the end of capability. If you're rebuilding a career, a body, or a life — the method here is transferable.
Start where you are. Build what you can.
The System
Two independent AI engines cross-check each other; a council of models decides. Every part earns its place — nothing runs that can't be explained.
Prices, news, filings, social sentiment, on-chain metrics, options, insider flow, and supply-chain mapping — pulled by 115 scheduled jobs.
Engine A (fundamentals), Engine B (technicals + cross-verify), Engine C (macro & sentiment) — a council split across A/B/C roles.
No single model is allowed to be right alone. A second engine independently re-derives the key numbers; divergence is surfaced, not hidden.
English to Telegram, Chinese to Feishu — every report carries a compliance line. WATCHLIST_ONLY by architecture.
One model can be confidently wrong.
Two engines catch it. A council of three decides.
The orchestrator. Runs fundamental, macro, and sentiment reasoning; schedules and sequences the entire nightly pipeline on macOS (Apple Silicon).
The independent check. Runs technical and flow analysis and cross-verifies Engine 1's numbers — the single most defensible feature in the system.
The Multi-Model Council
Fallback chain: GLM-5.2 → GPT-5.5 → DeepSeek V4 Pro → Qwen3 14B (local, free)
From Dusk to Dawn
All times US Eastern (ET). The system runs across the full US trading day and after-hours — collecting, reasoning, cross-verifying, and voting — so the verdict is ready when you are. The flagship tasks are highlighted.
Evening · Pre-market build-up
The most detailed scheduled report of the night: B-group panorama, position-by-position deep dive, A-group macro, C-group full list, D-group crypto, news & impact, valuation & technicals, a signal board, catalysts, and a data-quality check.
Kelly-criterion position sizing turns conviction into a number — then a clear BUY / HOLD / WATCH per holding.
Multi-engine deep analysis per ticker: Engine A (fundamentals + valuation + insider) → Engine B (technicals + cross-verify) → a co-authored Synthesis. 8 sections, ≤1,200 words, ending in BUY / HOLD / SELL + a target price. How it works ↓
Full-market hidden-gem discovery. Scans 8 sources and hard-filters out OTC, meme, penny, and micro-cap names. Five-factor scoring — Surge Intensity, Catalyst Backing, Supply-Chain Chokepoint, Market Blindness, Sector Alignment — sorts survivors into three tiers.
Night · Deep reasoning & the vote
A five-layer peel across 8 sectors — upstream (raw materials → components → equipment) and downstream (contract manufacturing → end brands). Each layer gets a bottleneck story (“who’s got who by the neck”) and a company deep-dive: moat, financials, valuation, buy/sell. Plain language, no jargon.
Per-holding bull case vs bear case, ≤3 points each, scored X/10 — the tension made explicit.
Covered call, protective put, bull call spread — or simply wait. A concrete structure for each position.
The three-model vote for contested calls: R1 DeepSeek proposes → R2 GLM-5.2 reviews → R3 GPT-5.5 breaks the tie → a single BUY / HOLD / SELL with a confidence score out of 100.
Overnight · Real-time watch
Four passes — 10:30 AM, 12:00 PM, 1:30 PM, 3:00 PM ET — tracking real-time B/C/D group prices, news, sentiment, and a 6-factor signal board. Force-pushes an alert the moment a holding moves ±4% / ±6%.
Dawn · The wrap-up
The most comprehensive daily read: B-group panorama, the C-group full list (49 tickers), crypto, sentiment divergence, oversold-bounce picks, next-session recommendations, and a data-quality report.
BTC network metrics, the Fear & Greed index, and per-coin sentiment — the crypto side of the morning brief.
Schedule reflects TMC v1.6.1 · US Eastern time · For research only · WATCHLIST_ONLY
Deep Analysis
A single model gives you one answer with no second opinion. TMC makes two independent engines derive the same numbers — and prints the disagreement instead of hiding it.
Confidently wrong, silently.
Disagreement is a feature, not a bug.
DeepSeek V4 Pro makes the opening call on the contested ticker.
A second model reviews the proposal, challenges the weak points, and revises.
The tiebreaker returns a single BUY / HOLD / SELL with a confidence score out of 100.
One model can be confidently wrong. Two engines catch it. A council of three decides.
No single vendor’s model bias — or outage — silently defines the whole analysis. WATCHLIST_ONLY.
How It Compares
Commercial platforms, open-source frameworks, and the DIY ChatGPT-and-spreadsheet approach. Here's where TradingMapClaw actually stands.
| Capability | TradingMapClaw | Commercial platforms | DIY ChatGPT |
|---|---|---|---|
| Monthly cost | ~$55 budget cap | $1,000s / seat | $20–200 |
| Cross-model verification | Yes — dual-engine | Rarely exposed | No |
| Fully auditable pipeline | Yes — every job | Black box | Ad-hoc |
| Automated & scheduled | 115 jobs nightly | Yes | Manual |
| Model fallback chain | 4-tier chain | Vendor-locked | No |
| Trade execution | None (by design) | Often included | No |
| Runs on your own hardware | One Mac mini | Cloud SaaS | Your choice |
WATCHLIST_ONLY: TradingMapClaw produces research only. It does not route orders or execute trades — and that boundary is intentional.
Skill Packs
The reusable engineering behind the system — each pack standalone, environment-agnostic, OpenAI-compatible, and MIT-licensed. Distilled from 93+ internal SKILL modules.
Track multi-model LLM spend against a monthly cap; GREEN / AMBER / RED status straight from a plain usage log.
One model answers; a second independent model verifies the key numbers before you trust the output.
Wraps scheduled jobs, auto-fixes common failure modes, retries once, and reports exactly what happened.
A pre-publish checklist — compile, freshness, schema, non-empty — that blocks bad output before it ships.
Tries a primary LLM, then falls back through an ordered backup chain on error, timeout, or rate-limit.
Versions your prompts and lints them — and their outputs — against your own house-style rules.
A 4-level pluggable fetch pattern — direct → alternate → library → cache — for any data field.
These seven are distilled from 93+ internal SKILL modules that run the live system.
These are part of the running system and not all individually for sale yet.
Products
Tutorials and reusable skill packs distilled from the system, plus hands-on consulting. Checkout goes live shortly.
Stand up a budget-disciplined, multi-model research pipeline from zero — the exact stack, guardrails, and fallback logic that keeps it cheap.
How two independent engines cross-verify each other — the Maker/Checker pattern, the council vote, and the quality gate that blocks bad output.
The full playbook: 115 jobs, 499 scripts, cron recovery, prompt governance, and budget engineering — everything that runs one system single-handed.
Seven standalone, environment-agnostic patterns — Budget Watchdog, Maker-Checker, Cron Recovery, Quality Gate, Model Fallback, Prompt Governance, Data Fallback. OpenAI-compatible, MIT-licensed.
Live From The System
Real report types, redacted to sample form. Every item ships with a compliance line.
B-group panorama, C-group full list, crypto, sentiment divergence, oversold-bounce candidates, and a data-quality report — the most comprehensive daily read.
Scans 8 sources, hard-filters OTC/meme/penny/micro-cap, then tiers candidates: High-Conviction, Emerging, Speculative Surge.
Upstream and downstream mapping across sectors — who has who by the neck — with chokepoint scoring in plain language.
DeepSeek proposes, GLM-5.2 reviews, GPT-5.5 breaks ties — final BUY/HOLD/SELL with a confidence score.
Samples for illustration · For research only · Not investment advice · WATCHLIST_ONLY
Get Access
Card and PayPal via a merchant of record that handles tax and receipts; crypto checkout going live shortly.
Instant, automated delivery via Lemon Squeezy — the merchant of record handling tax and receipts. Buy, and the files land in your inbox.
Pay in USDT or BTC via NOWPayments, with automated delivery. Wallets are whitelisted and pre-configured — going live once verification completes.
The Model
Nightly research at $55/month generates the raw material.
Every fix, playbook, and design decision becomes a tutorial or skill pack.
Buy the blueprint, or bring me in to audit and map your own build.
Which keeps running, keeps improving, keeps generating more to teach.
Writing
Technical, first-person, story-driven. New posts weekly.
The fallback chains, the guardrails, and the incident that nearly blew the budget.
SubstackDesigning an adversarial quality gate so divergence surfaces instead of hiding.
MediumWhat building a production system looks like from where I sit.
FAQ
One Hand · One Bag · One System