Research & Architecture

Yashraj Motwani

I go deep on the models everyone else treats as black boxes.

Five years at IIT Kharagpur, then R&D at EXL building NLP systems for Energy, Healthcare, and Travel clients. I've fine-tuned LLMs when fine-tuning wasn't trendy. Built intent clustering systems, NER pipelines, and retrieval architectures that actually worked at scale. The difference between a demo and a deployment is in the details — and I've spent years learning those details.

At ContextJet, I bring research depth to production problems. GraphRAG, hybrid retrieval, custom embeddings — I build the systems that make AI answers trustworthy, not just plausible.

Highlights

B.Tech + M.Tech, IIT Kharagpur

NLP R&D at EXL (Energy, Healthcare, Travel)

Research at Sungkyunkwan University, Seoul

Research at Johannes Gutenberg University, Mainz

Based in Bengaluru, India

Expertise

LLM Fine-tuning

Custom models with LoRA, Q-LoRA for domain-specific tasks. Not just prompting — actual model optimization.

RAG & Retrieval

State-of-the-art retrieval with ColBERT, SPLADE, hybrid search. Benchmarked, not guessed.

NLP Pipelines

Intent clustering, NER, summarization, question answering. Production-grade, not research-grade.

Selected Work

Enterprise Hierarchical RAG System

System architecture, LLM optimization, deployment

Architected production-scale RAG platform handling millions of documents across 160+ client deployments. Hierarchical document ingestion, multimodal search with image analysis, integrated mobile chatbot. Optimized for 100K+ requests/second throughput.

Nexo GraphRAG Engine

GraphRAG architecture, retrieval optimization

Engineered knowledge-graph-powered RAG combining Neo4j with vector embeddings for context-aware retrieval. Entity extraction, relationship mapping, multi-hop reasoning. Sub-second query latency at scale.

Unsupervised Intent Clustering

Architecture, clustering algorithms, visualization

Built transcript analysis system using GPT-3.5, LLaMA 2/3 embeddings with K-Means, DBSCAN, HDBSCAN for hierarchical intent clustering. Deployed for Energy, Healthcare, Travel clients.