Software engineer focused on backend systems and applied AI/ML.
BS Computer Science student at FAST NUCES, Karachi. Expected graduation: 2027.
I build backend and AI systems. My research focuses on recommendation reranking.
18 merged pull requests across 13 organisations.
| Organisation | Contribution | Merged PRs |
|---|---|---|
| VS Code C/C++: portable remote process selection | #14592 | |
| pprof: Windows browser launching and tool-path parsing | #1023, #1025 | |
| Warp: correct buffer aliasing in FEM shape optimisation | #1879 | |
| Candle: Symphonia 0.6 audio decoding compatibility | #3931 | |
| Native SDK: custom HTTP transports, minidump flags and bounded ELF build-ID parsing | #1987, #1911, #2055 | |
| compiler-builtins: NaN sign preservation and obsolete feature-gate removal | #1239, #1304 | |
| Wrangler: portable absolute module names without basename collisions | #15440 | |
| EUI: migrate a filter component to a function component | #9841 | |
| 4diac IDE: validate multiple adapter connections | #2944 | |
| Eventyay: POST-only state changes and consistent form buttons | #5542, #217 | |
| BCC: correct OpenZFS probe arguments in zfsslower | #5554 | |
| Migrate a community-library browser test to Playwright | #26888 | |
| Executable examples for dummy forecasting and classification catalogues | #11077 |
- ModelGate: OpenAI-compatible LLM gateway with provider failover and explicit streaming failure handling. TypeScript, Fastify, PostgreSQL and Redis.
- Walkz: AI-assisted PR review that verifies suspected regressions with reproducers on exact base and head commits.
- HookRelay: signed webhook delivery with persistent retries, dead letters and replay. TypeScript, PostgreSQL and BullMQ.
- Devonoma: GitHub activity dashboard with signed webhook verification and idempotent event storage. Next.js and PostgreSQL.
- Feasible Rerank: constrained recommendation reranking and the failure modes of QUBO penalty encodings.
- Budget Tune: hyperparameter search compared at equal measured CPU cost.
- Green Rerank: measured training and serving costs in recommendation pipelines.


