Backend systems that hold up
when AI sits on top of them.
I build the distributed infrastructure that AI features actually run on — Django & FastAPI services, Celery/Redis task pipelines, and Retrieval-Augmented Generation systems on PostgreSQL + pgvector. B.Tech in AI & ML, JNTUH 2025.
I'm a backend engineer who got curious about what happens when you put a real LLM pipeline behind a production-grade distributed system — not a notebook demo, a system with queues, workers, and failure modes.
Most of my work lives in Django/DRF and FastAPI, talking to PostgreSQL through carefully profiled queries, offloading anything slow to Celery workers behind Redis, and shipping it in Docker containers on AWS. The newer half of my work adds a layer on top: chunking documents, generating embeddings, storing them in pgvector, and serving answers through LlamaIndex + Gemini — without letting any of it block a request thread.
4+ months interning at Elevate Labs and CodSoft taught me the part nobody puts in a course: read the existing system carefully before you touch it, write the test before you trust the fix, and measure the improvement instead of assuming it.
Everything below is something I've shipped with, not just read about.
- Refactored legacy SQL queries and ORM logic in a production CMS, cutting average query execution time by 15%.
- Designed OAuth 2.0 authentication flows, hardening every external API endpoint against unauthorised access.
- Wrote pytest suites that reduced the critical bug backlog by 20% within the first sprint.
- Rated Best Performer (5-star) for code quality and turnaround on debugging.
- Built a Contact Management System with full CRUD using Python and SQLite, following MVC architecture.
- Added structured validation and exception handling, eliminating an entire class of data-integrity bugs.
- Shipped modular, PEP-8-compliant code — every pull request merged without revision.
A 5-container microservice that tracks flight prices around the clock and answers policy questions with a real RAG pipeline — not a toy chatbot bolted onto an API.
- 5-container architecture — FastAPI, PostgreSQL/pgvector, Redis, Celery worker, and Celery Beat orchestrated with Docker Compose, deployable with one command.
- RAG policy engine — chunks and embeds flight-policy documents with LlamaIndex + Gemini into pgvector, answering questions with cited sources and relevance scores.
- Non-blocking by design —
/track-flightsreturns 200 OK instantly; flight search and AI deal analysis run in background workers. - Rate-limited Amadeus integration — intelligent
asyncio.sleep()backoff prevents 429s across 50+ concurrent routes. - Zero-touch monitoring — Celery Beat runs a daily 08:00 sweep, compares live prices to saved targets, and fires Twilio SMS + Gmail alerts on its own.
AI-powered Django CMS. Django Signals trigger Celery tasks on every post save, generating SEO metadata and embeddings in the background so search works on meaning, not just keywords.
Django/DRF API built on ModelViewSets and DefaultRouters, cutting boilerplate by 50% and eliminating N+1 queries via select_related, profiled with Django Debug Toolbar.
Async FastAPI microservice with strict Pydantic validation and auto-generated OpenAPI/Swagger docs — built to bootstrap fast without losing type safety.
Harvested 1,000+ records from AJAX-heavy pages with Selenium and BeautifulSoup, with rate-limiting to respect anti-scraping defences and fully automated ingestion.
- Oracle Cloud Infrastructure (OCI) Foundations Associate2026
- HackerRank REST API Certification2026
- Data Statistics & Analytics (Python) — CITD, Govt. of India2024
- The Complete Python Pro Bootcamp — Udemy2025