Open to opportunities

Gaurav Kasat

@gauravkasat

Message

I build production LLM, RAG, and automation systems that improve healthcare and enterprise workflows.

Zimbabwe

Message

What I'm looking for

I'm looking to build and deploy high-impact LLM, RAG, agentic AI, and automation systems that improve real-world workflows.

At Risa Labs, I build medical AI systems that automate order processing, patient-data retrieval, and diagnosis workflows. My NAR stack increased untouched orders by 20% and saved annotators an average of three hours per day.

I developed Neo4j knowledge graphs and a dual-layer persistent knowledge base for patient similarity and chatbot-driven clinical queries. I also fine-tuned Qwen3-4B-Instruct for medical oncology to 83% accuracy, within 1% of a closed-source model.

Previously at Sprinklr, I trained and optimized LLMs using LLaMA, Mistral, and Phi architectures, and built RAG, summarization, actionable-insights, and account-research capabilities. My work improved in-house NLP accuracy by 6% and supported dynamic taxonomy customization across clients and industries.

I enjoy turning machine learning research into resilient, deployable systems using Python, PyTorch, Docker, Kubernetes, cloud infrastructure, and data pipelines.

Experience

Work history, roles, and key accomplishments

Current

Senior ML Engineer

Current

Risa Labs Inc.

Jan 2025 - Present (1 year 7 months)

Implemented a NAR stack solution to automate order processing, resulting in a 20% increase in untouched orders and saving annotators an average of 3 hours per day. Developed and deployed knowledge graphs using Neo4j to power patient similarity algorithms and serve as the backend Q&A engine for a chatbot, enabling dynamic patient data queries.

NAR Stack Neo4j Qwen3 4B Instruct RabbitMQ LLM Docker AWS Kafka Grafana Kubernetes

Senior Data Scientist

Sprinklr

Jan 2024 - Jan 2025 (1 year)

Trained large language models from scratch using architectures such as LLaMA, Mistral, and Phi, achieving high accuracy tailored to Sprinklr's use cases. Optimised LLMs using pruning, quantization, distillation, and speculative decoding, reducing costs and latency while preserving accuracy.

LLM LLaMA Mistral PHI Pruning Quantization Distillation Speculative Decoding RAG NLP Python PyTorch TensorFlow SQL MongoDB Qdrant Transformers fastAPI Pandas NumPy Jenkins Hugging Face GCP LangGraph

Data Scientist

Sprinklr

Jan 2022 - Jan 2024 (2 years)

Developed an advanced data annotation framework with noise reduction techniques, reducing processing time from days to hours. Developed and deployed generic sentiment analysis models, enhancing accuracy for 100+ languages, including low-resource ones, by over 35%.