-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Soham edited this page Jul 25, 2026
·
3 revisions
A survivability-first quantitative research & decision system, applied to algorithmic trading (India + US).
Debate before you build. Prove before you risk. Preserve the Core, always.
The canonical documentation is version-controlled in the repo's
docs/ folder. This wiki is just a map to it:
- Methodology — the LLM Council & two-layer screening (the interesting part).
- Philosophy — goals, the Dual Barbell, survive-don't-predict.
- Roadmap — the gated phases and their exit gates (incl. the Phase-3 grounded Q&A layer, "ARIA Chat").
- Decision log — every council verdict, dated.
- Concepts & methods — the quant/ML/systems methods, built vs planned.
- Safe agentic development — how an AI coding agent builds this without ever touching real money.
How ARIA differs from the typical "AI trading bot": LLMs debate the design; they never place trades. Paper-first, no leverage, nothing goes live until a strategy is proven honestly (deflated Sharpe). See the README.
Status: Phase 0 complete → entering Phase 1 (validating strategy edge on free platforms before building anything custom). Live progress: the ARIA Roadmap board — what's built, in progress, next, and deliberately deferred.