Kaushik Nishtala

About Kaushik Nishtala

Lead Data Scientist with > 5 YOE building scalable AI solutions across healthcare, IT, and early-stage startups - turning messy, high-stakes data problems into production systems that ship, scale, and stick.I build end-to-end AI platforms spanning policy optimized agentic systems, layout-aware document understanding, representation learning & embedding models, anomaly detection, and LLM fine-tuning & alignment (SFT/RLHF-style optimization), with emphasis on offline/online evaluation, reliability, model risk controls, and MLOps across AWS/Azure/GCP.To deep dive into my work, side projects, and technical notes: https://sosayskaushik.dev

Experience

1. Lead Data Scientist

   HiLabs

   Mar 2022 — Present · Washington, DC, US

   Contracts Intelligence - NegotiateAI: Architected an LLM/VLM-powered negotiation assistant that extracts and compares contract clauses and rate tables against playbooks/templates with source-grounded deltas, achieving 85–92% precision/recall on change detection- DocuDelta: Productionized a post-hoc contract comparison service that performs structure reconstruction, clause alignment, and semantic diffing to generate audit-ready change logs, reducing review effort from ∼20 hours to ∼3 hours per contract- DocumentAI: Trained and deployed a custom Swin Transformer (DONUT encoder)+ Faster R-CNN (FPN) layout model for bounding-box chunking, achieving 96% mAP, and improved table/image-heavy retrieval with ColPali-style multimodal search- Clinical Analytics Chat Assist (KG-RAG + Text-to-SQL)- Delivered a GenAI clinical Q&A system that answers encounter-level and longitudinal questions using knowledge-graph hierarchical RAG and a self-correcting Text-to-SQL pipeline with constrained decoding for safe, reliable queries on regulated clinical data- Directed ML anomaly detection design, enhancing healthcare provider directory accuracy and saving $10M in claims processing- Improved data pipeline efficiency using Hadoop and Spark, and migrated infrastructure to AWS to bolster downstream operations- Established robust, secure infrastructure using self-hosted LLMs and asynchronous SQS process- Developed a scalable system for processing provider contracts, focusing on entity extraction to enhance claims processing- Engineered a document processing pipeline with LayoutLM for extracting entities from diverse media types- Served as primary technical liaison to clients like Elevance Health, translating complex algorithms into actionable business insights.

Education

• Northeastern University

  Master of Science, Data Science