AI Engineer · Full-Stack · Building production LLM systems | RAG · PyTorch · Next.js · FastAPI | Open to global opportunities
I build AI systems that ship — not prototypes, not notebooks, production products with real users. Currently: CS (AI) undergraduate. Already working like a professional engineer. What I've built: → Afflixor — full-stack AI affiliate platform with LLM recommendations, L1+L2 caching, 500+ concurrent users on free-tier infra → GPT-style transformer model built from scratch in PyTorch — every attention head written by hand, 98k training steps, live at minigpt-98.streamlit.app → Enterprise RAG pipeline — FastAPI + ChromaDB + provider-agnostic LLM adapter pattern → WooCommerce AI plugin — submitted to WordPress.org, supports Claude, GPT-4, Gemini, LLaMA with zero switching cost → Contributor to Twenty (open-source CRM, 25k+ GitHub stars) My stack: AI/ML: PyTorch, LangChain, n8n, RAG, ChromaDB, LLM APIs (Claude, GPT-4o, Gemini, LLaMA) Backend: FastAPI, Node.js, Express, PostgreSQL, Redis Frontend: Next.js, React, Flutter, TypeScript Infra: Docker, Vercel, Railway, DigitalOcean 400+ LeetCode problems solved. I'm based in India and open to remote roles globally — internships, contracts, or full-time in AI engineering. Best way to see what I build: minigpt-98.streamlit.app
Grras Solutions Pvt. Ltd.
Jun 2025 – Aug 2025 · Jaipur
Integrated Gemini LLM into internal tools, handling prompt engineering, response parsing, and error handling. Reduced API latency by approximately 20% through query profiling, N+1 elimination, and targeted caching.
B.Tech, Computer Science (AI)
2023 – 2027
CGPA: 8.18 / 10
Production-deployed full-stack affiliate marketplace built with Next.js, Node.js, PostgreSQL, Redis, Groq, Prisma, Railway, and Vercel. Implemented affiliate tracking, CMS features, a three-stage AI recommendation pipeline, semantic retrieval with HuggingFace embeddings and Qdrant, LLaMA-based product reasoning, and L1/L2 caching.
Implemented and trained a 50M-parameter transformer from scratch using PyTorch, including multi-head self-attention, causal masking, positional embeddings, GELU FFN, mixed precision, cosine learning-rate scheduling, gradient clipping, and checkpoint resumption. Deployed a Streamlit web app with controllable generation.
Designed a modular FastAPI backend with RAG pipelines and ChromaDB for context-aware enterprise knowledge-base responses. Built provider-agnostic adapters and REST APIs for multi-step AI workflows and an enterprise dashboard.
WordPress plugin submitted to WordPress.org that generates product titles, descriptions, tags, and metadata using Claude, GPT-4, Gemini, and LLaMA. Supports BYOK model selection, multiple providers and sub-models, and custom prompt editing.