Headline

AI Research Engineer @ BMW Techworks India | LLMs · Agentic AI · RAG · Fine-Tuning · AI Strategy | Ex-Analyttica

About

I'm an AI Engineer at BMW TechWorks India, deep in the trenches of building production-grade Generative AI systems for automotive intelligence. My days revolve around Agentic AI architectures, fine-tuning Large Language Models (7B+ parameters using LoRA/QLoRA), and training domain-specific NLP models that actually work in the real world — not just in notebooks. At BMW, I get to experiment with cutting-edge LLM research and ship it. Whether it's orchestrating multi-agent systems with LangChain and CrewAI, optimizing retrieval pipelines with vector databases, or squeezing performance out of transformer models, I'm hands-on with the full stack — from research prototype to production deployment. Previously at Analyttica Datalab, I architected something I'm particularly proud of: a custom Copilot system that integrated three core AI engines (processing, inference, and recommendation) to automate complex workflows. This wasn't just a chatbot — it was an intelligent orchestration layer that reduced processing time by 30% and became a core product feature. I also engineered a RAG-powered recommendation engine that drove a 70% lift in customer engagement, and fine-tuned Llama-3.2 to jump conversational AI accuracy from 70% to 90%. What drives me? Pure curiosity and an almost obsessive need to understand how things work under the hood. I don't just implement — I question, I experiment, I break things, and I rebuild them better. I take complete ownership of my work, from that first research paper I read to the final production deployment at 2 AM. The field moves fast, and I move with it — constantly learning, constantly shipping. Core Technical Stack: → Agentic AI & Multi-Agent Orchestration (LangChain, CrewAI, LlamaIndex) → LLM Fine-Tuning & Alignment (Llama, Mistral, GPT-4, LoRA, QLoRA) → RAG Systems & Vector Databases (Pinecone, MongoDB, hybrid retrieval) → NLP Model Training & Optimization → Production ML Pipelines (PyTorch, Transformers, MLOps) Currently exploring: Efficient fine-tuning methods, autonomous agent behaviors, and the intersection of LLMs with automotive intelligence. Always open to connecting with fellow builders, researchers, and anyone else who geeks out on making AI actually useful.

Location

Bengaluru

AI stack

github copilot

hugging face

langchain

langsmith

openai

vLLM

Work experience

AI Research Engineer at BMW TechWorks India, Jan 2026 - Present · 10 mos, Bengaluru, Karnataka, India

Data Scientist at LEAPS by Analyttica, Jan 2024 - Nov 2025 · 1 yr 11 mos, Bengaluru, Karnataka, India, • Led technical development of "Copilot" — an AI assistant integrating 3 core engines (processing, inference, recommendation), enhancing enterprise level decision making and analytics • Designed and implemented an enterprise-scale KPI Knowledge Graph integrating heterogeneous data sources, enabling intelligent analytics and automated decision-making workflows for business stakeholders • Fine-tuned Llama-3.2 (7B parameters) for enterprise conversational AI using LoRA techniques, custom datasets, and rigorous evaluation frameworks • Architected and deployed a production-grade RAG (Retrieval-Augmented Generation) system powering a recommendation engine, driving 70% increase in customer engagement through contextual, AI-generated suggestions • Built and optimized AI pipelines using GPT-4o and LangChain for contextual data generation, automated content creation, and intelligent workflow automation • Implemented multi-turn conversational AI bot with MongoDB session management, enabling persistent context across complex user interactions • Developed API integration router using React Agent architecture with Pinecone vector embeddings, achieving sub-200ms retrieval latency for 10,000+ document corpus • Delivered 5+ high-impact Proof of Concepts (POCs) leveraging few-shot learning, function calling, and fine-tuning strategies. Technologies: Python | LangChain | LlamaIndex | GPT-4o | Llama-3.2 | Pinecone | MongoDB | PyTorch | Hugging Face | Streamlit | AWS | Knowledge Graphs | Neo4j

Data Analyst at Shadowfax, Sep 2023 - Jan 2024 · 5 mos, Bengaluru, Karnataka, India

Data Science Trainee at AlmaBetter, Aug 2022 - Aug 2023 · 1 yr 1 mo, Bengaluru, Karnataka, India, - Gained skills in Analytical tools, Visualization tools, Machine Learning, Statistics, Hypothesis testing, Natural Language Processing, and teamwork. - Worked on Projects like building a recommender system, Retail Sales Prediction, Covid-19 Tweet Sentiment Analysis, and more. - Secured a place in the top 3% of students in the cohort of 300+ students, and got selected for the prestigious Star Student Program, which allowed me to connect with top-performing peers, and receive mentorship from experienced professionals.

Education

Bachelor of Technology - BTech, Computer Science, Rajasthan Technical University, Kota

Public links

github: https://github.com/HarshJain41

medium: https://medium.com/@harshjain_

harshjain41.github.io: https://harshjain41.github.io/harshjain-portfolio