Open to opportunities

Aayush Kumar

@aayushkumar3

Message

Data Scientist specializing in domain LLMs, RAG, and efficient inference.

India

Message

What I'm looking for

I’m looking for a role where I can ship enterprise Generative AI with domain LLM fine-tuning, RAG, and document intelligence—prioritizing measurable accuracy, factual grounding, and lower latency through efficient inference and production MLOps.

I’m a Data Scientist with 3+ years of hands-on experience designing and shipping AI/ML systems across Generative AI, LLMs, NLP, and Deep Learning. I specialize in domain-specific LLM fine-tuning, retrieval-augmented generation, document intelligence, and transformer optimization—focused on measurable accuracy and latency gains in enterprise production.

At IBM (Data Scientist), I led domain-specific LLM fine-tuning for enterprise GenAI using Gemma and Qwen on 50+ research papers, improving performance by 15–25% while reducing training time by 30%+. I also built QA, summarization, classification, and RAG pipelines for air-pollution intelligence, and created a document intelligence platform processing 10K+ documents to generate hierarchical knowledge graphs and mind maps.

I bring additional leadership experience through commissioned service as a Flying Officer with the Indian Air Force, and I co-invented a patent-pending adaptive inference framework for rank-sliced LoRA adapter execution to improve efficient LLM serving throughput. My work also emphasizes reusable training/evaluation frameworks, hallucination/relevance/faithfulness metrics, and production-grade pipelines that deliver faster experimentation and more reliable outputs.

Experience

Work history, roles, and key accomplishments

Data Scientist

IBM, Indian Software Labs

Jul 2025 - Dec 2025 (5 months)

Spearheaded domain-specific LLM fine-tuning and built QA, summarization, classification, and RAG pipelines for an enterprise air pollution intelligence system. Developed a document intelligence platform with hierarchical knowledge graphs and contributed a patent-pending adaptive inference framework for efficient LLM serving.

LLM Fine tuning LLM Retrieval Augmented Generation (RAG)Document Intelligence Knowledge Graphs LlamaIndex PEFT Adaptive Inference Evaluation Metrics (Hallucination Relevance Faithfulness)Instruction Tuning

Software Engineer (AI)

Jio Platforms, EasyGov

Aug 2022 - Jun 2025 (2 years 10 months)

Integrated GPT-based generative AI and NLP models for content generation, summarization, OCR, and information extraction across government datasets and portals. Led AI platform development and built secure REST APIs and speech-to-text pipelines to automate grievance processing and document generation.