🎓 ECE Undergraduate @ RUET '28, Bangladesh Backend Engineer (Python · FastAPI · PostgreSQL) | AI Systems Builder

I build production-quality backend systems and AI-powered applications — real systems with real design decisions, load-tested, containerized and shipped through CI.

📄 Resume · 💼 LinkedIn · ✉️ topukumar538@gmail.com

Multi-agent incident investigation system built on a LangGraph DAG

Code · Live Demo

• 5-node LangGraph DAG — log analysis and timeline extraction run in parallel and join at root-cause inference, over three temperature-tuned LLM instances (0.7 chat / 0.3 RAG / 0.1 postmortem) isolating generative, grounded and deterministic work
• Verified against 3 documented production incidents — GitLab 2017, Cloudflare 2019, AWS 2020 — replayed through the pipeline, asserting both root-cause category matches and domain-specific terms drawn from the published postmortems
• Stateless HMAC-SHA256 session tokens carrying a version field — bumping one database column invalidates every issued token, with the server storing none of them
• Restart-transparent state — FAISS stores persisted to disk under per-user, per-session paths and conversation state mirrored to PostgreSQL, so summary buffers and recent messages survive a reconnect
• Confidence-aware pipeline — halts and asks clarifying questions when the evidence score drops below 60%
• 123-test GitHub Actions pipeline runs on every push and deploys only when tests pass, replacing manual redeploys

FastAPI · LangGraph · LangChain · FAISS · PostgreSQL (async SQLAlchemy) · Groq (Llama 3.3 70B) · Docker · Neon

Production-style backend with a slot-based explore–exploit ranker

Code · Live Demo

• 99% success rate at 100 concurrent users (~800ms p95) — load tested with Locust from 10 to 500 users, and root-caused the failure at 500 as database connection pool exhaustion
• No single category dominates the feed — capped at 3 of the top 5 slots per page by a slot-based ranker using softmax sampling at temp=0.7 with 10% random injection
• Recommendations follow current behaviour, not old activity — interaction weights (view=1, like=3, save=5) decayed on a 30-day constant and recomputed for every user in hourly batches
• Zero repeats across the entire feed — all pages built in one pass against a shared seen-set, seeded so pagination stays stable at a full 20 posts per page
• Admin blocks apply in one request instead of up to 7 days — user state re-checked in the database on every authenticated request rather than trusted from the JWT
• 58-test GitHub Actions pipeline automating deployment to Hugging Face Spaces

FastAPI · PostgreSQL · SQLAlchemy 2.0 · JWT + OTP Auth · APScheduler · Docker · Pytest · Locust

Also: Regulated Power Distribution System — STM32-based embedded team project at RUET, built with @Md-Yousuf-2723 and Anindita Sarkar.

Languages

Backend & Databases

AI / ML

Cloud & Infrastructure

Tools

Rajshahi University of Engineering & Technology (RUET) — B.Sc. in Electrical & Computer Engineering Expected Oct 2028 · CGPA 3.51 / 4.00 Coursework: Data Structures & Algorithms, Object-Oriented Programming, Database Systems

• ✅ Machine Learning Specialization — DeepLearning.AI / Andrew Ng (Coursera) · Dec 2025
• ✅ CS50x: Introduction to Computer Science — HarvardX (edX) · Oct 2024
• ✅ CS50P: Introduction to Programming with Python — HarvardX (edX) · 2024

• 📌 400+ DSA problems solved across LeetCode, GeeksforGeeks and Codeforces — arrays, trees, graphs, binary search, dynamic programming
• ☁️ Working through AWS fundamentals
• 🌱 Preparing for open-source contributions in the CNCF ecosystem (Prometheus, OpenTelemetry)
• 💼 Open to Summer 2027 Software Engineering internships