Shreyan Das
| # | Project | Stack | Role | Description | Status |
|---|---|---|---|---|---|
| 1 | Munim.ai | Python · FastAPI · llama.cpp | Co-founder | Offline-first bank reconciliation engine matching statements against ledgers. Exports a four-tab audit-ready Excel report with written justifications, running all inference on-device. Cut LLM involvement to roughly 1 row in 12 via a five-stage deterministic pipeline. | Shipped |
| 2 | Solis — Agentic OS | Fedora Linux · Agent Runtime | Team Project | A Linux distribution where planning and task execution are first-class system services. Agent runtime and permission surface designed around OS primitives — processes, files, packages — so every agent action stays inspectable and reversible. | In Progress |
| 3 | Live-Market Algo Trading Platform | Python · FastAPI · React | Solo Build | Research platform trading real-time crypto data against a simulated wallet across three strategies, judged on live performance not backtests. Risk-invariant execution engine covered by 368 tests, including a no-look-ahead verification suite and live monitoring dashboard. | Shipped |
| 4 | Mini-orch | Python · Tkinter · llama.cpp · GGUF | Solo Build | Zero-setup desktop app bundling llama.cpp and GGUF models — no Ollama, no API keys needed. Provider layer routes local, Ollama, OpenAI and Anthropic models by tier and task. Supports manual and self-planning fan-out multi-agent modes with budget, routing, and permission gates. | Shipped |
| 5 | Spoken Micro-Learning App | Next.js · TypeScript · Tailwind · On-device LLM | Product Design & Prototype | Phone-native app that captures spoken curiosity, researches it on a scheduler gated on device state, and delivers a 90-second spoken lesson when the user opens a doomscroll app. On-device pipeline with quantized 2–4B LLM, local speech with barge-in, and spaced repetition. | Prototype |