cv

Applied AI Scientist

Microsoft

Finetuning Small Language Models with RLHF and custom reward policies for data security compliance.

• Finetuning Small Language Models using RLHF and custom Reward policies for data security compliance
• Automated finetuning workflow using MLFlow and tested SLMs using RAGA's package
• Built a custom Model registry system using MLFlow, GCP and PostgreSQL with RBAC Control and model serving capabilities using Kubernetes

Machine Learning Engineer (Project Associate)

Indian Institute of Science, Bangalore (AIREX LAB)

ML research in collaboration with DRDO on defense surveillance and LLM serving infrastructure.

• Developed ML models in collaboration with DRDO for radar intercept point classification, improving accuracy by 25%
• Developed an advanced defense surveillance system combining YOLOv10, LIDAR, and SLAM technology
• Built high-throughput LLM REST API services using FastAPI, Kubernetes and Apache Kafka with 100ms average inference time
• Developed a RAG pipeline using LangChain for a course-oriented knowledge base from a Milvus Vector Database

Founding Engineer (Machine Learning Engineer)

Zenteiq.ai (Incubated at IISc, Bangalore)

Founding engineer building LLM serving infrastructure, multiagent pipelines, and backend services for an AI startup incubated at IISc.

• Delivered POCs with IndiaAI, Artpark, and Garett Motion, securing a central government fund of $10 Million for India AI Mission
• Built high-throughput LLM Engine using FastAPI, Kubernetes and load balancers processing 1k+ concurrent requests/sec at 200 tokens/sec
• Architected end-to-end face recognition tracking system with LLM-powered search, reducing latency by 60% and improving matching accuracy by 90%
• Engineered user engagement tracking service using Google Pub/Sub and BigQuery with in-memory caching and complete ETL pipeline
• Designed Knowledge Graph infrastructure using Neo4j for a personalized learning platform
• Developed multiagent MCP and RAG pipeline using Claude, LangChain, CrewAI and LangGraph over Milvus, Timescale DB and MongoDB

Research Intern

National Institute of Technology, Kurukshetra

Research on medical image segmentation and autonomous robotics under Dr. Gaurav Verma.

• Conducted renal blood vessel segmentation using DeepLabv3+, achieving ROC of 72%, 10% above prior benchmark
• Designed and developed an autonomous Rubik's Cube solver using DIP techniques and the CFOP algorithm
• Curated and annotated a Rubik's Cube dataset for face detection, achieving 98% accuracy

Research Intern

University of Central Florida

Research on pose estimation, Vision-Language Models, and molecular property prediction under Dr. Yogesh S. Rawat and Dr. Shruti Vyas.

• Applied Vision Transformers and CNNs for pose estimation on NTU-120 Skeleton Points Dataset, improving accuracy by 10% to 95%
• Curated datasets for chemistry-based task predictions using Vision Language Models (Llama Adapter, BLIP, CogVLM)
• Leveraged LLMs and VLMs (BLIP2, LLaVA 1.5) to surpass chemistry benchmarks by over 20% — Accepted at NeurIPS 2025
• Assessed LLM-based molecular property prediction (Phi4, Deepseek, GPT) improving baselines by 27% — Under review at ICML 2026

Software Intern

Microsoft — Cloud & AI

Developed and tested cloud solutions using C# and Azure integrations.

• Designed and implemented solutions using the C# framework
• Implemented integration test cases and bug fixes to ensure high-quality code
• Integrated PowerBI template app solutions for Azure, streamlining automated deployment by 60%
• Implemented integration between Marketplace, Appsource, and Solution Center, reducing deployment services by 62.5%

B.Tech

National Institute of Technology, Kurukshetra

Electrical Engineering

• Data Structures and Algorithms
• Operating Systems
• Mechatronics
• Soft Computing

JEE Mains All India Rank 15244

National Testing Agency

Top 1.3 Percentile among 1.1 Million students

Microsoft Engage Mentee

Microsoft

Selected out of 20,000 applicants

Smart India Hackathon National Finalist

Smart India Hackathon

Top 5 teams out of 500+ — Hyderabad, India

First Position — PIXAR (Digital Image Processing Competition)

NIT Kurukshetra

First place out of 200+ participants

Bronze Medal — Kaggle Forecasting Mini Course Sales

Kaggle

Bronze medal for Kaggle notebook

SiP: Structure-informed Prompting for Molecular Property Prediction

COLM 2026

Under review. Deepan Adak*, Dr. Yogesh Singh Rawat, Dr. Shruti Vyas.

MolVision: Molecular Property Prediction with Vision Language Models

39th NeurIPS 2025

Accepted. Deepan Adak*, Dr. Yogesh Singh Rawat, Dr. Shruti Vyas.

OvarianNet-CA: A Hybrid Cross-Attention Ensemble Model Approach Using MixTransformer and EfficientNet

4th OPJU International Technology Conference, 2025

Accepted. Deepan Adak*, Srushti Sonawane, Dr. Gaurav Verma.

Cross-Modal Intelligence for Autonomous Manipulation: HFC-YOLOv11 Feature Correlation Networks and B-Spline DE-MPC Integration for Rubik's Cube Solving

International Journal of Intelligent Robotics and Applications, Springer

Under review. Deepan Adak*, Aryan Kumar, Dr. Gaurav Verma.

DeepSegPro: Renal Vasculature Segmentation in HiP-CT Imaging with Attention-Enhanced Deep Learning for Medical Assessment

Biomedical Signal Processing and Control, Elsevier

Under review. Deepan Adak*, Dr. Gaurav Verma.

LesionLensPro: Skin Cancer Detection using an Ensemble Approach with EfficientNetV2 and ResNet

8th International Conference on Micro-Electronics and Telecommunication Engineering

Accepted. Deepan Adak*, Dr. Gaurav Verma.

Even More Fruits

Machine learning fruit classification full-stack web application

• Developed ML model using MobileNetV2 achieving 95% accuracy, 95% Precision and 93% F1 Score
• Built full-stack web application with React.js frontend and Flask backend
• Deployed on Microsoft Azure for scalability and reliability