🧠 AI Engineer | 🔬 Building AI from scratch — not just using it | 📍 Bangalore, India
Deep in the weeds of transformers, tokenizers, and training loops. I build models and systems from the ground up to actually understand how they work — then wrap them in something a user (or a recruiter) can actually click through.
Currently building multi-agent AI systems and grinding on NLP and computer vision while interviewing for ML/AI roles.
🚗 ENFORE v2 — Safe Multi-Modal Autonomous Driving (M.Tech Thesis, IIT Bhubaneswar) Multi-modal perception pipeline for safe autonomous driving — fuses camera and LiDAR data via HydraNet, with QP-based output correction for runtime safety enforcement. Submitted to RV 2026.
🤖 Multi-Agent Research AI System Coordinated multi-agent pipeline — ReAct search & reader agents (LangGraph) backed by custom Tavily/BeautifulSoup tools, LCEL writer & critic chains for self-reviewed report generation. Live Streamlit demo.
🏥 Medical Chatbot — RAG with LangChain & AWS RAG pipeline grounded in the Gale Encyclopedia of Medicine — LangChain, Pinecone vector search, Flask backend, deployed via Docker + CI/CD.
📝 Small Language Model — Built From Scratch Decoder-only transformer (GPT-style): custom BPE tokenizer, multi-head self-attention, positional embeddings, nucleus sampling — no high-level model libraries.
🧾 Multimodal OCR Pipeline — Insurance Document Digitization CNN + metadata fusion model to classify scanned insurance IDs — PyTorch, 10-epoch training with LR scheduling.
🎯 SharpenAI — Interview Prep App React/Vite app for structured interview prep, built for the Polaris Fellowship 2026 (shortlisted Round 3). Live
📈 Market Intelligence & Trend Analysis Engine Google Trends time-series engine across 20+ categories — seasonal decomposition and COVID-era demand analysis to guide product-growth strategy.
Transformers CNNs NLP BPE Tokenizer Multi-Head Attention RAG LangGraph ReAct Agents Pinecone BeautifulSoup
📧 tariqhusainkhan01@gmail.com | 💼 LinkedIn
"I don't just call
.fit()— I build the pieces that make it work."