Sidharth Shambu

  1. Bothos - An autonomous code-maintainer agent with custom agentic harness for solo open-source maintainers. It can investigate issues, explore repositories, modify code, run validations, upgrade dependencies, and open pull requests with proposed fixes. The agent loop and tool harness are built from scratch in Go using pi-agent-sdk.

  2. DockNudge - A small Go service that watches Docker's live event stream and sends focused alerts for OOM kills, crashes, unhealthy containers, restart bursts, and unexpected stops. Supports 8 notification backends with cooldowns, expected-stop filtering, and reconnect handling. Useful when you want failure visibility without running a full monitoring stack.

  3. Multi-Tenant GPU Platform (genv) - Built a shared GPU platform on 2Ɨ NVIDIA L40S GPUs using genv for soft VRAM slicing, per-user Docker containers, and XFS project quotas. Added watchdog enforcement and SSH-based container access so multiple users can safely run LLM and training workloads on the same host. Technical writeup in my blog.

  4. Google Summer of Code @ LibreCube - Implemented CCSDS Unified Space Link Protocol framing, error detection, and mux/demux on ESP32-class hardware for CubeSat radio links. Also built a C++/Python test harness to validate protocol cases. Full report: shambu.bearblog.dev/googlesoc-21-librecube

  5. upi-intent - A framework-agnostic TypeScript library that takes a standard upi://pay?... URI and generates app-specific deep links for Google Pay, PhonePe, Paytm, and more. Includes URI validation, typed APIs, and a QR fallback for desktop flows. Useful when you want one payment button that opens the right UPI app.

  6. Quantimind - A therapist-client analytics dashboard built as a freelance project. A LangGraph pipeline processes therapy transcripts into moods, triggers, cognitive distortions, and behaviors, while a Next.js dashboard shows session trends, alerts, and drill-downs for clinicians.

  7. znlp - A local AI terminal assistant that turns natural-language instructions into shell commands. Commands require confirmation before execution, and inference runs through local Ollama models so prompts and terminal context stay on the user's machine. Also available as a cross-platform npm package.