Vibhas Singh

About Vibhas Singh

Currently leading Generative AI & ML initiatives for an industry-scale AI-assisted medical coding platform processing 100M+ medical record pages per month. 6+ years of experience productionizing classical ML and deep learning systems in healthtech — evolving from feature-engineered models (XGBoost/CatBoost) to transformer-based architectures and now to GenAI-native, retrieval-augmented, agentic AI systems. Driving the evolution from traditional ML pipelines to LLM-powered, agentic systems — building autonomous medical coding and verification agents using LLMs, structured generation, fine-tuning, distillation and reasoning-optimized prompting to improve human-in-the-loop workflows. Architecting and deploying billion-scale inference systems for clinical document intelligence — leveraging SOTA inference servers (e.g. vLLM), mixed precision (FP16/BF16), torch.compile, async inference, and distributed multi-GPU training for high-throughput, low-latency production workloads. Designing retrieval-augmented and multi-stage reasoning systems — combining dense embedding models (encoder-based, domain-adapted transformers), ANN search, re-ranking pipelines, synthetic data generation, and custom evaluation frameworks to improve reasoning consistency and taxonomy alignment (e.g, ICD-10 mapping). Leading large-scale domain adaptation and continual pretraining of transformer models with MLM and custom objectives; fine-tuning LLMs & VLMs using QLoRA/LoRA for OCR-free structured extraction, long-context understanding (8K+ tokens), and medical reasoning tasks. Building modular, scalable ML platforms for clinical document AI — spanning text, layout, vision, and multimodal transformers (ModernBERT, BioMedRoBERTa, LiLT, LayoutLM, GTE, Qwen-VL etc.), enabling document understanding, layout parsing, visual-NER, assertion detection, semantic search, and structured information extraction. Passionate about building robust evaluation loops, synthetic data pipelines, scalable deployment templates, and high-performing ML teams — mentoring data scientists while pushing the boundary of applied clinical AI.

Experience

1. Principal Data Scientist

   Optum

   Mar 2025 — Present · Mumbai, IN

   Bigger stakes - and even bigger impact.Driving last mile efficiencies in medical coding workflows using AI.Leading a team of 7 data scientists.

Education

• The LNM Institute of Information Technology

  Bachelor of Technology (B.Tech.), Electronics and Communication Engineering

  2016 — 2020