Abhishek Chaudhary

About Abhishek Chaudhary

How I Can Be of Value to You• Design and deliver production-ready GenAI solutions that solve real-world business challenges and drive measurable results• Build end-to-end Agentic RAG systems that combine retrieval, reasoning, and automation• Deploy LLMs in secure on-premise environments for organizations prioritizing privacy and performance• Fine-tune LLMs for domain-specific use cases, enhancing relevance, accuracy, and efficiency My Areas of Expertise• Generative AI | Agentic RAG Systems | On-Premise & Cloud LLM Deployment• LLM Fine-Tuning | Prompt Engineering | Retrieval-Augmented Generation• Python Programming | Forecasting | Advanced Modeling | Anomaly & Drift Detection• Strong Engineering & Mathematics Background | Data-Driven Decision Making• Cloud & Hybrid Infrastructure | Scalable AI Deployment

Experience

1. Data Scientist

   TartanHQ

   Jul 2024 — Present · Gurugram, IN

   PolyGPT - Developed an Agentic RAG system with Pinecone VectorDB and RBAC for policy-related queries- Improved accuracy and efficiency in policy development through intelligent information retrieval using Pinecone vector DB along with reranker.On-Premise LLM Deployment - Deployed Meta LLaMA 3.1 8B with vLLM, LiteLLM, and Open-WebUI for secure access- Ensured data privacy by keeping all AI interactions within a closed VPC. In-House Policy Document Parsing Service - Built a scalable microservice using multiple OCR and parsing tools with intelligent fail-safes- Enabled secure, high-throughput policy document extraction in markdown format. AI Inferno: AI Monitoring & Stability Assurance - Created a closed-loop AI monitoring system for real-time anomaly detection- Improved AI reliability and security by mitigating risks like prompt injection. Assessment AI - Automated MCQ generation from policy documents using GPT-4o- Reduced manual workload for policy managers, ensuring consistent policy awareness testing. Fine-Tuning of Open-Source Model for Specific Use Cases - Used Unsloth and Transformers to fine-tune open-source models for domain-specific tasks- Improved model performance and adaptability for specialized business applications. Accuracy Improvement of Payslip Data Extraction - Enhanced data extraction accuracy by refining existing pipelines and integrating advanced OCR techniques- Reduced errors and improved automation efficiency in processing payslip information.

Education

• Krishna Institute of Engineering & Technology

  Bachelor of Technology - BTech

  2013 — 2017

• National Institute of Technology, Tiruchirappalli

  Master of Technology - MTech

  2020 — 2022