Nishant Kumar
@nishant7264
Postgrad Researcher at AI-NLP-ML Research Lab | IIT Patna
Bangalore, Karnataka, India
Nishant Kumar is a Data Scientist and AI Researcher currently pursuing an M.Tech in Artificial Intelligence at IIT Patna. He has professional experience developing multi-agent architectures at Omnissa and forgery detection systems at Signzy. His research focuses on trustworthy radiology impression generation using preference optimization. Nishant holds a B.Tech in Electrical Engineering and possesses expertise in machine learning, NLP, and computer vision.
Experience
Postgrad Researcher
AI-NLP-ML Research Lab | IIT Patna
Led NLP healthcare research under Dr. Sriparna Saha, developing a framework that reduced hallucinations in AI-generated radiology impressions by 45%. Implemented and optimized advanced RL techniques (DPO, PPO, RLOO). Trained 3 specialized reward models that enhanced safety compliance and factual accuracy. Analyzed 10,000+ radiology reports to enhance impression generation.
Data Scientist
Omnissa (formerly VMware EUC)
Built a natural language driven workflow generator using a multi-agent LangGraph architecture that decomposes queries, resolves resources identifiers via APIs, and constructs UEM workflows from user intent. Integrated with Freestyle Orchestrator, Omni Chat, and VMA Agent to support automated workflow building and deployment. Built a robust testing framework to ensure reliability and scalability.
AI Engineer
Signzy
Developed an advanced forgery detection system with 96% accuracy. Enhanced Aadhar masking API accuracy to 98% by training a custom YOLO model. Improved DL Extraction API field accuracy by 19% through LLM integration. Created unified Auto-Extraction API wrapper combining 8 KYC document services using Gen AI.
Data Scientist
Silwalk
Engineered AI-powered meeting solution on Zoom Web SDK with HLS streaming. Implemented real-time transcription and summarization with 94% accuracy. Built a speaker identification system and automated topic generation. Developed RAG-based Q&A system for meeting queries.
Junior Data Scientist
Productize Tech
Designed an Automatic Number Plate Detection system with 95% accuracy using YOLO, Deep SORT, and PaddleOCR. Executed real-time Firebase integration for plate data updates.
Education
Indian Institute of Technology Patna
M.Tech
Artificial Intelligence
Muzaffarpur Institute of Technology
B.Tech
Electrical Engineering
Licenses & Certifications
Programming with Python
Coding Ninjas
Data Structures and Algorithms
Coding Ninjas
Machine Learning
Coding Ninjas