Backend and ML engineer. B.Tech CSE at VIT Bhopal and BS Data Science at IIT Madras, studied side by side. Final year.
Portfolio: porfoliosample-zeta.vercel.app | LinkedIn | Email
Most students build for the happy path. I build for edge cases, race conditions, and partial failures.
I build backend services that stay correct under concurrency and keep running when a dependency fails, and ML systems whose results I can defend with an experiment log. I am looking for SDE, backend, or ML-focused engineering roles at teams that care about system design and reliability.
| Project | What it is | Stack | Links |
|---|---|---|---|
| Placement Portal | Campus hiring platform for students, recruiters and admins. 53 documented REST endpoints, stateful JWT logout (JTI blacklist), Redis caching, Celery jobs, and fallbacks at every dependency. | Flask, Vue 3, SQLAlchemy, Redis, Celery | Code / Live |
| Healthcare Appointment Manager | Clinic booking where two patients cannot take the same slot (compare-and-set holds, idempotency keys, race tests). Gemini triage and summaries with a deterministic fallback, email outbox, Google Calendar sync. | FastAPI, PostgreSQL, Redis, Celery, Docker | Code / Live |
| InventoryPro | Inventory SaaS where stock changes only through valid order-state transitions. Edge-verified JWT, Recharts dashboards, PWA. | Next.js, TypeScript, Prisma, PostgreSQL | Code / Live |
| Project | Result |
|---|---|
| Smart MCQ Solver | 0.75084 MAP@3 (Kaggle private) on a 2,000-row STEM dataset. Fine-tuning lost to zero-shot LLM logit scoring, and the repo documents all 16 approaches. |
| FinSight AI | Multi-agent RAG over financial documents: LangGraph, BM25 plus FAISS with rank fusion, cross-encoder reranking, guardrails, RAGAS evaluation. |
| Toxic comment classifier | 0.826 Macro F1, top ~150 of 600+ teams, classical ML only. Stacking beat voting; SMOTE tested and rejected. |
| Solari demand forecasting | 970K rows, 1,115 hubs, LightGBM, XGBoost and CatBoost behind a FastAPI service with CI. |
| Hugging Face semantic matcher | FlashText, KeyBERT and BGE embeddings in tiers; 50% lower inference memory with one shared model instance. |
hld-distributed-systems-portfolio holds five runnable services, each also its own repo.
| Service | Pattern |
|---|---|
| URL shortener | Base62 codes, LRU cache with single-flight, JWT |
| Rate limiter | Five algorithms, Redis or in-memory, fail-open or fail-closed |
| Real-time chat | WebSockets, Redis Pub/Sub backplane, transactional outbox |
| News feed | Push, pull and hybrid fan-out, cursor pagination |
| Ticket booking | Optimistic locking, expiring holds, saga with compensation |
These run locally; payments and some storage are simulated.
- Concurrency. Compare-and-set state changes and idempotency keys instead of heavy table locks.
- Resilience. If Redis, a queue or an LLM is down, the core flow degrades to a simpler path instead of failing.
- Experiments. I keep negative results, because they usually explain the final number.
- Documentation. Design notes and READMEs state what is simulated and what is production.
- Languages: Python, C++, C, Java, TypeScript, JavaScript, SQL
- Backend: FastAPI, Flask, SQLAlchemy, Celery, Redis, PostgreSQL, SQLite, JWT, WebSockets
- ML and AI: PyTorch, scikit-learn, LightGBM, XGBoost, Hugging Face, Sentence-Transformers, LangChain, LangGraph, FAISS, Weights & Biases
- Tools: Git, Docker, Linux, GitHub Actions, Vercel, Railway, Fly.io
- B.Tech CSE, VIT Bhopal (expected 2027)
- BS Data Science and Applications, IIT Madras (expected 2027): Diploma in Programming, Diploma in Data Science, two advanced certificates
- NPTEL Elite certificates (Machine Learning, Marketing Analytics), HackerRank SQL (Advanced), Coursera Supervised ML, AWS Educate badges
- MPOnline (Govt. of MP and TCS JV) AI/ML and software engineering internship, May to Jul 2026
- 950+ day GeeksforGeeks streak, 200+ LeetCode problems