I work on backend reliability and systems correctness: tracing race conditions, state leaks and memory bugs, then shipping small, tested fixes upstream.
Snapshot · Upstream work · Systems · Projects · Experience · Contact
I'm a recruiter — the 20-second snapshot
- Target roles: Summer 2027 SDE / Backend / ML internships
- Education: B.Tech, Computer Science & Business Systems at PDEU — 9.61/10, expected May 2028; Diploma in Computer Science — 9.78/10
- Upstream engineering: Contributions to Celery, Django, Microsoft Agent Framework and Open Food Facts, including fixes for concurrency, ORM state, workflow isolation and memory handling
- Systems: C++20 limit order book and matching engine — 278× throughput improvement on a cancel-heavy workload; tested with ASan, UBSan and TSan
- Industry experience: Django backend internship; database indexing and ORM tuning reduced average API response time by 25% under production load
- Achievements: JPMorgan Chase Code for Good 2026 winner; Google The Big Code 2026 Round 2; Amazon ML Challenge 2025 top 2,500 of 30,000+
- Competitive programming: Codeforces Expert (max 1738, current 1704); LeetCode (2207)
- Contact: [email protected]
I'm an engineer — show me the technical work
Start with the upstream bug investigations: a Redis chord race in Celery, a deferred-field overwrite in Django, failed-superstep state leaking across Agent Framework workflow runs, and a traceback/frame reference cycle. Then explore the C++ matching engine, including its lock-free SPSC ingestion queue and sanitizer-driven debugging.
I'm a maintainer — what is my contribution style?
I focus on reproducing the failure, understanding the invariant that was broken, keeping the patch scoped, and adding regression tests. I value review feedback and prefer a smaller, well-tested fix over a broad refactor when the smaller change solves the underlying problem. In Django, maintainer review shrank my patch to a change confined to DeferredAttribute, and the fix is better for it.
$ whoami
namraa@upstream
focus backend reliability · systems correctness
interests concurrency · memory · shared state · distributed systems
approach reproduce → isolate root cause → minimal fix → regression tests
upstream Celery · Django · Microsoft Agent Framework · Open Food Facts
builds C++ matching engine · async backend systems · cloud applications
status open to Summer 2027 SDE / Backend / ML internships
| 278× | 25% | 9.61/10 |
|---|---|---|
| matching throughput gain | API latency reduction | B.Tech CGPA |
I contribute fixes and features to production repositories, with particular interest in correctness under concurrency, state lifecycle, memory behavior and failure handling.
| Failure | Repository | Root cause / fix | Merged PR |
|---|---|---|---|
| Redis chord callback could publish twice | Celery | Made completion detection and dispatch an atomic Lua-scripted claim; added unique claim tokens and ownership-validated cleanup | #10998 |
| Deferred fetch overwrote existing values | Django | Prevented FETCH_PEERS from overwriting already-loaded or locally modified values; kept the change scoped to DeferredAttribute |
#22177 |
| Failed workflow superstep leaked staged writes | Microsoft Agent Framework | Discarded pending state after failure or cancellation across core, declarative and foundry-hosting packages | #8819 shipped in 1.22.0 |
| Hard-timeout traceback/frame reference cycle | Celery | Replaced ineffective traceback clearing with exc.__traceback__ = None and added regression/smoke tests |
#10493 shipped in 5.7.0 |
| Redis result consumer deadlocked | Celery | Guarded a redundant UNSUBSCRIBE path that re-entered a non-reentrant lock |
#10497 shipped in 5.7.0 |
| Synthetic compaction summaries disappeared | Microsoft Agent Framework | Reworked before_run reconciliation so summaries survive subsequent agent runs |
#7944 shipped in 1.19.0 |
| Certificate expiry during verification | Celery | Added an explicit expiry check before certificate verification | #10510 |
Browse all my merged pull requests on GitHub →
Selected contributions in other collaborative repositories
A few examples beyond the core upstream projects:
| Engineering area | Repository | Selected merged work |
|---|---|---|
| Distributed rate limiting and authentication | LegalEase | Redis-backed rate limiting #441 · strict JWT/API-key separation #398 |
| Application security | Eventra | CSRF protection and enforcement #8150 |
| RAG security | pdf-qa-bot | SSRF hostname validation #403 · remove unsafe FAISS deserialization #388 |
| Async runtime and concurrency | AegisGraph Sentinel · pathfinder-ai | synchronize shared runtime state #1054 · replace timer-based mutex polling with queued waiters #1388 |
| Webhook verification | pathfinder-ai | verify Clerk webhook signatures using Svix #2268 |
A price-time-priority matching engine supporting limit and market orders, partial fills, cancellations and multiple instruments.
- 278× throughput improvement on cancel-heavy load: replaced a map/deque order book with a sorted-vector plus intrusive-list design, improving cancellation to O(1); measured 13.7K → 3.81M ops/s.
- Allocation efficiency: a chunked freelist object pool reached a 99.99% hit rate, adding 22–49% throughput and up to 5× lower variance. Profiling-led tuning added another 15%; regressing experiments were reverted.
- Concurrency correctness: designed a lock-free SPSC ingestion queue with acquire/release ordering and a single-writer matching thread. ThreadSanitizer exposed a real memory-ordering race, which I fixed.
- Validation: caught a silent-drop flaw in my own benchmark and added
submitted = processedchecks; completed 490 test runs under ASan, UBSan and TSan.
C++20 CMake GoogleTest Google Benchmark ASan UBSan TSan
TeamSense · Async task pipeline for HR analytics
- Built an asynchronous Celery + Redis pipeline to move compute-heavy feedback processing out of the request cycle, reducing synchronous processing load by 60%.
- Designed a multi-entity PostgreSQL schema for employees, meetings and feedback; containerized the stack with Docker for reproducible environments.
Django DRF Celery Redis PostgreSQL Docker React
Personal Cloud Assistant · Serverless AWS application
- Built a notes-and-files backend with AWS Lambda and API Gateway, DynamoDB and S3 presigned download URLs.
- Secured access with Cognito JWT authentication and least-privilege IAM roles.
- Added CloudWatch custom metrics for operation counts, request duration and errors behind an admin-only dashboard; shipped a React 19 frontend with Amplify CI/CD.
AWS Lambda API Gateway DynamoDB S3 Cognito CloudWatch React
ProfiLens · Career-preparation platform
- Built a normalized, indexed PostgreSQL schema for user progress and gamification data, with role-based access control and session-scoped permissions.
- Added a background ReportLab pipeline for certificate and resume generation.
Django DRF PostgreSQL ReportLab
| Experience | Details |
|---|---|
| Django Backend Intern · Techaronic IT Solutions | Ahmedabad · Jun–Aug 2024. Tuned database indexes and ORM queries, reducing average API response time by 25% under production load; implemented authentication/authorization workflows; deployed and debugged AWS EC2 services under live traffic. |
| B.Tech — Computer Science & Business Systems · Pandit Deendayal Energy University | Gandhinagar · 9.61/10 · Expected May 2028 |
| Diploma Engineering — Computer Science · Government Polytechnic Ahmedabad | Ahmedabad · 9.78/10 · Aug 2022–Apr 2025 |
AWS services I've worked with
| Service | What it does | Where it appears in my work |
|---|---|---|
| Amazon EC2 | Virtual servers for running applications | Deployed and debugged backend services under live traffic during my internship |
| AWS Lambda | Runs code without managing servers | Serverless backend for my Personal Cloud Assistant |
| Amazon API Gateway | Creates and manages API endpoints | Routes requests to the assistant's Lambda functions |
| Amazon DynamoDB | Managed NoSQL database | Stores application data for the assistant |
| Amazon S3 | Object storage for files and data | File storage and presigned download URLs |
| Amazon Cognito | User sign-up, sign-in and token-based identity | JWT-based authentication for the assistant |
| AWS IAM | Controls identities, roles and permissions | Least-privilege permissions for backend resources |
| Amazon CloudWatch | Logs, metrics and operational monitoring | Custom metrics for operation counts, latency and errors |
| AWS Amplify | Frontend hosting and application delivery | React frontend deployment and CI/CD |
- Backend reliability & distributed systems — async execution, task queues, retries, failure handling and state consistency.
- Systems programming & performance — concurrency, memory behavior, data structures, profiling and measurable optimization.
- AI-powered applications — practical ML, retrieval-augmented generation and agent workflows.
- Open-source infrastructure — debugging real-world issues, writing regression tests and collaborating with maintainers.
Achievement details
- JPMorgan Chase Code for Good 2026: Winner; built a production-ready solution for a nonprofit partner.
- Google The Big Code 2026: Qualified for Round 2.
- Amazon ML Challenge 2025: Top 2,500 among 30,000+ participants.
- Open Source Hackathon: Winner; delivered production-ready features under time constraints.
- Hackout 2025: Finalist on technical merit.
- Certifications: Oracle Cloud Infrastructure Generative AI Professional; Oracle Data Science Professional.
