CKCK YONG
Start a project
Back to Portfolio
PUBLIC REPOSITORY

O.D.I.N

Zero-cost decision-support system built solo, running high-stakes personal and business decisions through four independent AI reasoning passes before arbitrating one clear execution blueprint.

Visit Project
Omni-Dimensional Intelligence Node (O.D.I.N), a private, zero-cost decision-support system built solo to run high-stakes personal and business decisions through four independent AI reasoning passes — arbitrated into one execution blueprint — built because a single chat LLM's hedged, risk-averse answer isn't the same as genuinely adversarial, structured reasoning. Format A structured intake — Core Objectives, Known Constraints, Raw Narrative — feeds three independent Gemini API calls in sequence: the Quant estimates expected value across possible paths, the Strategist maps reversibility and positioning, the Behaviorist audits the narrative for sunk-cost, confirmation, and overconfidence bias. Once all three finish, a fourth Judge call receives only their completed outputs and arbitrates a synthesis, calling out exactly where they agree and where they collide. Every call runs live against real API status, and the JSON output is schema-enforced, so a completed session always returns exactly three sequenced next actions and a Mermaid diagram — never fewer, never vaguer. What it covers Built for genuinely high-stakes dilemmas with incomplete information — personal decisions, interpersonal risk calls, business or strategic trade-offs — not routine questions. The Strategist persona detects whether a given decision is competitive or non-adversarial and reframes its reasoning accordingly, so the same engine handles something like "should I say something" and something like "how do I price against fast-following competitors" without needing separate tooling. Why it works Standard chat LLMs converge on one generalized, risk-averse answer because that's what a single reasoning pass optimizes for. Splitting the problem into four independent calls — cold optimization, adversarial strategy, bias audit, first-principles arbitration — means the personas genuinely have different outputs to weigh against each other, rather than one model quietly agreeing with itself. Running entirely on free-tier Streamlit and Gemini keeps it sustainable at zero cost, which matters because this is a personal tool with no funding or revenue behind it, built and maintained solo.
O.D.I.N screenshot 1
O.D.I.N screenshot 2

Key Features & Details

  • Runs every decision through four independent reasoning passes — Quant/Strategist/Behaviorist/Judge — instead of one hedged LLM answer
  • Structured JSON schema enforces exactly 3 next actions and a Mermaid diagram on every session
  • Live progressive-disclosure status shows each real API call in flight (not a fake progress bar)
  • Built on Streamlit/Gemini API and Google Antigravity IDE — zero cost end to end