Full-Stack & AI/ML Engineer specializing in resource-efficient AI systems, agentic workflows, and end-to-end full-stack architectures. Experienced in maximizing LLM performance under strict budget and context constraints (AST-based context selection, token pruning).
NextGen Invent
Remote
– Architected and shipped production-grade full-stack applications using React.js, Django REST, and FastAPI, improving API response times by 35%.
– Built responsive Tailwind CSS UIs and RESTful API integrations across multiple client projects.
Modern Monk Labs
Remote
– Developed high-throughput FastAPI inference endpoints to serve deep learning models, handling up to 150 concurrent requests per second to a React.js front-end; integrated PostgreSQL pipelines for efficient storage and retrieval of AI-generated outputs.
VRARMR
Remote
– Built NL2SQL models (natural language → SQL); ran A/B tests on speech recognition improving multilingual accuracy to 76% ; automated workflows with LangChain agents.
B.Tech
7.1 CGPA
Python · AST · Dependency Graphs · LLM Evaluation
– Developed a repository-scale context selection engine for code LLMs that maps multi-file dependency graphs and applies token-budget-aware pruning, slashing prompt tokens from 35,270 down to 10,747 (a 69.5% context reduction).
– Built runtime tracer (sys.settrace()) to generate gold-standard execution graphs; achieved 79.6% precision on Requests library benchmark (249 nodes, 181 edges). Evaluating against BM25, semantic retrieval, and naive graph expansion baselines.
React · FastAPI · PostgreSQL · CoinGecko API · OCR · OAuth
– Engineered an end-to-end fintech dashboard integrating real-time market data pipelines, localized GST calculations, and an automated OCR bill scanner processing document uploads with 94% accuracy.
React.js · Node.js · TMDB API · PostgreSQL
– Full-stack streaming site with TMDB integration, responsive UI, user auth, and content delivery / ad management back-end — personal tool still in active use.
Engineered automated feedback workflows via SAP AI Core; benchmarked 3 Foundational models (Gemini, Mistral & GPT) in production; optimised enterprise prompt accuracy.
LangChain modules for enhanced RAG retrieval · Gradio demos for text-to-image AI models.