Working · Architected for productionPaxie — Agentic RAG Knowledge Assistant
Ask your company's documents in plain language, get cited answers
An R&D effort I led and built solo — a knowledge assistant that lets employees query company documents in natural language and receive AI-generated answers with source citations.
- Multi-pass Agentic RAG with automatic query reformulation (up to 3 passes) — vs. single-pass in most RAG apps
- Full role-based access control enforced on every vector search, with permission-gap detection
- Pre-index quality gate — low-quality documents are scored and rejected before ingestion
- Multi-provider AI — Claude, Gemini, and Azure OpenAI, switchable from an admin panel
- Built for self-hosted deployment: Docker Compose, nginx, TLS, GitHub Actions CI/CD
Next.js 16PostgreSQL + pgvectorPython FastAPIClaude APIPrismaDocker
Live demo coming soon
Built & working · Deployment-readyProject Instance — AI-Augmented Development Workflow Platform
Mission control for building software with AI
A platform that turns the AI-assisted development workflow into a managed product — one dashboard to plan, build, test, document, and ship — replacing scattered planning and handoff files. Conceived from lessons learned building Paxie, and built solo.
- Multi-role SaaS with full RBAC — Owner, DevOps, Developer, Tester, and read-only Viewer
- Quality gates that block a build phase until tests, docs, and open issues are resolved — with an audited owner override and a fix-and-re-verify issue loop
- GitHub integration — scaffolds new repos or adopts existing ones in safe pull-only mode
- Multi-environment management — Development → Staging → Production with a guided promotion flow
- Delivery-metrics dashboard plus an embedded read-only AI assistant answering questions from each project's own records
Next.js 16React 19TypeScriptPostgreSQLPrismaNextAuthDockerAzure AI Foundry
Live demo coming soon