Deep Research Agent, an autonomous research agent built on an agent-controller pattern. It breaks a question into ordered sub-tasks, searches the web with Tavily, and writes a report with citation tracking and contradiction detection. LLM providers are swappable through dependency injection.
0
13
MEGA AI, a multi-agent orchestration and evaluation backend. An orchestrator routes each query through RAG, critique, synthesis and meta agents, streams partial results over SSE, saves execution traces and runs an evaluation loop with adversarial robustness tests. Built with FastAPI, LangChain, PostgreSQL and Docker.
0
14
Inventory Ledger and Invoice Extraction. Barcode-driven parts custody for an office-supplies company: every part moves through an append-only ledger (purchase in, issue, install or return), so its current location and holder can be rebuilt from the event log at any time. Built with Flask and SQLite and deployed on the office network over Tailscale. For the same client I built an LLM pipeline that extracts vendor, GST number, line items and taxes from supplier invoices into clean CSV, and a Next.js B2B product catalog.
0
22
Custom Restaurant POS. A restaurant operations stack for a quick-service restaurant in Noida: touch-screen kiosk ordering with Razorpay payments, a kitchen display and order management, a customer queue screen and a CMS-backed website, running on-premise. The POS is offline-first, so orders queue locally and reconcile when the connection returns, and service keeps running through network outages. Built with TypeScript, Next.js, SQLite and Drizzle in a pnpm and Turborepo monorepo.
0
29