Hi, I'm Yashraj.
NLP Engineer & AI Researcher.
Co-founder at ContextJet.ai.

LLM Fine-tuning & Optimization

Custom models with LoRA, Q-LoRA for domain-specific tasks.

RAG & Retrieval Systems

State-of-the-art retrieval with ColBERT, SPLADE, hybrid search.

NLP Pipelines

Intent clustering, NER, summarization, question answering at scale.

Enterprise Hierarchical RAG System for Hardware MNC

Architected production-scale RAG platform handling millions of documents across 160+ client deployments. Built hierarchical document ingestion, multimodal search with image analysis, integrated mobile chatbot. Optimized Google Gemini and OpenAI endpoints achieving 100,000+ requests/second throughput with cost-efficient LLM routing.

Role: system architecture, LLM optimization, deployment

Nexo GraphRAG Engine

Engineered knowledge-graph-powered RAG system combining Neo4j graph database with vector embeddings for context-aware retrieval. Implemented entity extraction, relationship mapping, and multi-hop reasoning for complex queries. Deployed for enterprise knowledge management with sub-second query latency at scale.

Role: GraphRAG architecture, knowledge graphs, retrieval optimization

Unsupervised Intent Clustering System

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.

Role: architecture, clustering algorithms, visualization

LLM Fine-tuning Framework

Fine-tuned GPT-3.5, Llama 2, flan-t5 using LoRA, Q-LoRA, few-shot learning techniques for domain-specific generation across multiple industries.

Role: model optimization, custom training

Advanced NER Pipeline

Stacked BERT-CRF and reverse question answering models with iterative RoBERTa fine-tuning, achieving cutting-edge accuracy for entity extraction.

Role: pipeline design, model stacking

LangChain Competitive Analysis Tool

Built AI-powered analysis tool with Streamlit + LangGraph agent. Engineered robust RAG pipeline benchmarking ColBERT, SPLADE, DPR, BM25-BERT for optimal retrieval.

Role: full-stack development, RAG architecture

AI-powered Product Discovery Assistant

Built knowledge graph using NetworkX, RDF, Schema.org, migrated to Neo4j. Created LangChain query processor translating natural language to structured KG queries.

Role: knowledge graph design, query processing

Customer Sentiment Analysis (Bosch India)

Achieved 93.22% accuracy using BERT models and TF-IDF feature engineering. Built CSAT Django web app with automated scraping and real-time sentiment analysis.

Role: research, web app development