Dainwi Kumar.

Zero-friction polling app — create a poll in seconds, share the link, watch votes stream in live. Node.js/Express backend uses Redis Pub/Sub to broadcast every vote to all connected clients via Server-Sent Events; two dedicated ioredis instances handle commands and subscriptions separately. Redis TTL auto-expires polls, a Redis Set prevents duplicate votes per browser UUID, and poll data is stored as Redis Hashes with nanoid-generated short IDs. Proper routes → controllers → services architecture with PostgreSQL via Drizzle ORM as secondary store.

Full-stack job application tracker with a sortable/filterable TanStack Table pipeline for desktop and a tab-strip mobile list view; per-application prep checklists backed by Server Actions. Integrated Drizzle ORM with Neon PostgreSQL; used React.cache() for per-request query deduplication. Wired Resend email into a Vercel cron job for daily batched deadline reminders; rate-limited REST routes handle CSV bulk import up to 1 MB.

Batch resume screener that processes up to 20 PDF/DOCX files in a single run against a job description. Client-side polling streams per-candidate scores progressively into the UI as the async FastAPI backend scores each resume — live progress indicators replace a blocking spinner. One-click CSV export of ranked results.

Transforms any query into a citation-backed research report by orchestrating a multi-agent pipeline: query expansion, web search, content scraping, LLM ranking, and synthesis — processing 15–30 sources per session. Implements concurrent page fetching via HTTPX with MD5-based deduplication. Real-time SSE streaming across the full stack.

A Rust-based proxy that sits between autonomous AI agents and MCP servers, intercepting JSON-RPC 2.0 messages over stdin/stdout. Implements WAL-mode SQLite logging for full audit trails, TOML-based policy enforcement to allow/deny specific tool calls, and per-tool rate limiting via token bucket algorithm. Uses tokio::select! to handle child process lifecycle and prevent zombie processes.

An AI-powered code knowledge base that lets developers query an entire codebase in plain English with exact file citations. Built four features: semantic vector search, RAG chat with SSE streaming, file-similarity recommendations, and an agentic Q&A pipeline that decomposes complex queries into sub-tasks. Runs fully local with Ollama — zero API cost, zero data leaving the machine.