I'm a final-year Computer Science student at BIT Mesra (graduating May 2027), looking for SDE-1 / new-grad software engineering roles. I like the backend side of real-time products — sync protocols, storage, rate limiting, and the bugs that hide in shared state — and I fix those bugs in other people's code too: my patches are merged in Microsoft's Agent Framework and LibreDB Studio.

• 🔀 7 merged pull requests in open source, including fixes in microsoft/agent-framework — with more in review at pytest, Swift, Mermaid and Zulip
• 🚀 CollabEdit — a live Google Docs–style editor on a CRDT sync server I wrote in Python, with 250+ automated tests
• 🏆 70th of 8,500+ teams in the Amazon ML Challenge 2026 · 🥇 1st place at Hatch From Scratch (AR indoor navigation)
• 🧮 600+ DSA problems solved in C++ · LeetCode contest rating peaked at 1605

Real-time collaborative documents in the spirit of Google Docs — live cursors, offline editing, version history, full-text search and link sharing — built from scratch on FastAPI, Yjs/pycrdt, PostgreSQL and Redis.

▶ Try it live · Source · Design decisions (ADRs) · Roadmap

• Own sync server. FastAPI WebSockets speak the Yjs sync and awareness protocols on top of pycrdt (Rust Yrs): one room per open document, per-connection send queues, and roles enforced on every update. In production, edits reach other users in about 110 ms.
• Edit log + compaction. Every edit lands in an append-only Postgres log (batched every 500 ms) and is folded into snapshots under an advisory lock — opening a 10,000-keystroke document went from 218 ms to 17 ms.
• Self-built rate limiter. A token bucket in one Redis Lua script: atomic across API instances, in a single round trip.
• Careful sessions. Argon2id passwords, 15-minute JWTs, rotating refresh tokens with reuse detection, and one-time WebSocket tickets.
• Offline-first. The browser keeps an IndexedDB copy, and offline edits merge automatically when the connection returns.
• Tested and documented. 173 API tests against real Postgres and Redis (including Hypothesis convergence properties) and 79 web tests run in CI on every push; 10 architecture decision records explain the trade-offs.

I look for real bugs in projects I use, reduce them to a minimal reproduction, and fix them with regression tests. Some favourites:

• microsoft/agent-framework #8660 — is_type_compatible special-cased Any only as a target type, so workflow edge validation rejected executors declaring WorkflowContext[Any].
• microsoft/agent-framework #8648 — middleware type detection failed under from __future__ import annotations; postponed (PEP 563) annotations are now resolved before dispatch.
• libredb/libredb-studio #1121 — a single SELECT, AUTH or HELLO could silently move or re-authenticate the Redis connection shared by every later request; blocked at the provider boundary after verifying each command against live Redis 8.

✅ Merged — 7 pull requests in 4 projects

🔄 In review — 5 open

_{Both tables refresh daily from the GitHub API.}

Also: SQLAlchemy · Alembic · WebSockets · Socket.IO · Yjs / CRDTs · JWT · OAuth 2.0 · pytest · Hypothesis · Vitest · Render · Vercel

CS fundamentals: data structures & algorithms, operating systems, computer networks, DBMS

📫 adityaccds@gmail.com · LinkedIn — happy to talk about backend roles, real-time systems, or anything above.