2025 B.Tech AI/ML graduate specializing in Python backend engineering and applied AI — multi-agent LLM pipelines, RAG systems, async and distributed backend infrastructure, and production-style deployment. Based in Hyderabad, India.
I’m a 2025 B.Tech AI/ML graduate with a focus on Python backend engineering and applied AI. I like understanding how things actually work—not just at the surface, but the tradeoffs, failure modes, and reasoning behind technical decisions.
The intersection of AI and backend systems is where I want to be: multi-agent pipelines, RAG systems, async infrastructure, and systems where intelligence meets scale. I’ve independently shipped 8 projects, completed software engineering and research internships, co-authored an IHCI 2023 paper, and contributed 4 merged PRs to an open-source FastAPI project. I’m looking for roles where there’s always something deeper to figure out.
Route-based fuel stop optimizer built for the Spotter AI Backend Django Engineer assessment. Takes a start/finish US location, fetches the driving route via a single OpenRouteService API call, and finds the cheapest way to fuel up across 8,100+ real truck stations loaded into PostgreSQL. Candidate stations are narrowed to the route using vectorized NumPy haversine distance checks against the full polyline. Stop selection uses a dynamic programming approach — modeling the trip as a shortest-path problem over reachable stations within a 500-mile tank range, with a stop-penalty term that discourages excessive refueling stops while still reporting a fully reconciled, real fuel cost total. Deployed on Railway with a frontend for testing.
Automated benchmark comparing a graph database against 4 alternatives — Neo4j AuraDB, Memgraph, FalkorDB, and self-hosted Neo4j — across traversal, lookups, aggregation, and concurrent throughput. Measured p50/p95 latency across 100+ iterations per query and diagnosed a driver-protocol bug and a network-RTT-dominated latency floor on cloud free tiers.
Production-grade async BFS web crawler with dual fetch strategy — aiohttp for static sites, automatic Playwright fallback for JS-heavy sites. robots.txt compliance with async per-domain caching. 3-table PostgreSQL schema with background job tracking and duplicate detection. Containerized and published on Docker Hub.
Multi-agent LangGraph pipeline processing insurance PDFs end-to-end — classifies pages into 9 document types via Gemini vision, routes to specialized extraction agents. Exponential backoff retry logic, graceful JSON fallback, keyword-based mock layer for testing without API quota consumption.
Role-based finance dashboard backend with 3-tier RBAC (Admin, Analyst, Viewer) across 15+ REST endpoints. JWT authentication, bcrypt hashing, soft delete, dynamic filtering, and pagination. 33 passing pytest tests with dependency injection overrides and isolated test database.
RAG pipeline over 119K+ hotel booking records — chunked and embedded domain documents with sentence-transformers (MiniLM-L6-v2), indexed via FAISS cosine similarity, Mistral-7B for natural language Q&A. Hybrid retrieval combining vector search with deterministic analytics for revenue, cancellation rate, and lead-time queries. 4 FastAPI endpoints with background embedding on startup; analytical reports returned as base64 charts.
Horizontally scalable file sharing platform with NGINX load balancing across 3 FastAPI instances. Supports 2GB chunked uploads. Automated self-cleanup via asyncio background tasks. One-command deployment with Docker Compose.
Multi-criteria task prioritization system using Weighted Sum Model from MCDM theory with 4 prioritization strategies, dependency management with forward/backward handling and penalty multipliers. RESTful API with 100% pytest coverage.
First exploration of multimodal API integration — Cohere for NLP, Imagga for image recognition. Built async middleware and a responsive JS frontend. May be broken on free-tier hosting.
First attempt at RAG — LangChain, vector database for semantic search, LLM APIs with response streaming. Where I learned what retrieval-augmented generation actually means in practice.