IAR — Governance‑First AI Portfolio

An Elixir/OTP platform showcasing a portfolio powered by an AI chat assistant with continuous compliance, full traceability, and self‑improving agents.

Quick Start

• Dev stack: make dev (frees port 4000, starts db/OPA, detects host Ollama, runs migrations, boots Phoenix).
• Health: GET /api/health
• Chat: POST /api/chat with JSON { "message": "..." } and header x-org-id.
• Trace summary: GET /api/traces/:trace_id/summary.

Architecture (MVP)

• Runtime: Phoenix API, OTP supervisors, Oban jobs, Postgres + pgvector, OPA, Ollama.
• Orchestrator → Agents with plan/act/explain + decision records in audit.
• Memory: vector (RAG) + relational graph. Multi‑tenant via Postgres RLS.

Dev Tooling

• Scripts: bin/dev, bin/test, bin/stop, bin/port-kill.
• Make targets: dev, testc, phx, format, lint, dialyzer.
• CI: format, Credo, tests, Dialyzer, Sobelow.

Workflows

• Issue‑first: open an Issue (Feature/Bug/Chore templates). Intake creates a branch and a draft PR (labeled needs-HIL).
• PR template enforces Summary, Scope, Risk/Governance, Testing, and Checklist.

Links

• Contributor Guide: AGENTS.md
• Product Requirements: docs/PRD.md

Roadmap

• M2: RAG embedding pipeline; SpreadsheetAgent (CSV→XLSX); PDFAgent; HttpFetch.
• M3: OPA preflight in flows; HIL UI; trace viewer page.
• M4: ToolSmith agent generation; auto PR + HIL gate.
• M5: Observability (OTel, metrics); blue/green deploy docs.