SankalpBorse

AI/ML Engineer building agentic systems.

Projects

AI systems

These are projects across agentic support systems, retrieval, reinforcement learning, model compression, and multimodal analysis.

Main Project

Strategy-Based Emotional Support Agent

A strategy centric conversational Agent. That provides emotional support to the user through strategy focused communication.

DistilBERTLLM APIsRAIPythonPytorchTransformers

Open repository

Strategy-Based Emotional Support Agent

The system models emotional support as a decision-making process. It separates perception, session tracking, strategy prediction, response generation, privacy filtering, fairness monitoring, explainability, and safety governance so support conversations are more interpretable and controlled. User Input is combined with context, session state and user's emotion lable and this entire vector is feeded to a distil BERT Classifier that outputs one of the strategies (Advisement, Open Question, Interpretation..etc) This Strategy along with specified prompt along with context is provided to a LLM to generate a appropriate response.

GitHub

Major Project

RAG-Based Smart Kitchen Assistant

A stateful, voice-enabled cooking assistant with pantry-aware recipe retrieval and real-time cooking guidance.

FastAPIChromaDBSQLiteGeminiGroqWhisper

Open repository

RAG-Based Smart Kitchen Assistant

The assistant can Retrieve recipies from a RAG Database (specificaly created for cooking recipies through online datasets). Assistant can provide step by step guidance to the user through voice in the creation process. Its keeps track of pantry, user preferances and user's food related allegies. It can customize the recipies based on user preferances and Allergies.

GitHub

Deep RL

AI Agent Plays Snake Game

A Deep Q-Learning agent trained to play Snake Game Autonomously

PythonPyTorchPygameDQNRL

Open repository

AI Agent Plays Snake Game

The project uses replay memory, Bellman optimization, exploration-to-exploitation learning, reward shaping, and multiple play modes to show how an agent learns spatial strategy through trial and error. And learns to play in a contrrolled environment.

GitHub

Model Compression

Self-Pruning Neural Network

Applied a differentiable gating mechanism to a ANN that lets a neural network prune weak connections during training.

PyTorchCIFAR-10GatingSparsityJupyter Notebook

Open repository

Self-Pruning Neural Network

Each weight is paired with a learnable gate score. A sparsity penalty pushes unnecessary gates toward zero, producing compact models while preserving accuracy across CIFAR-10 experiments.

GitHub

Multimodal AI

Real-Time Human Behavior Analysis

A webcam-based dashboard for live cognitive state estimation from emotion, gaze, iris, and pose signals.

FastAPIMediaPipeFEROpenCV

Open repository

Real-Time Human Behavior Analysis

The system streams browser frames to FastAPI, runs open-source models, computes TESI over a rolling window, and displays emotion, gaze, posture, cognitive state, and history in real time.

GitHub

Capability Stack

Detailed technical skills.

The stack covers model development, data engineering, deployment, frameworks, and the fundamentals needed to build reliable ML products.

Programming & Core CS

PYPythonSQSQLCCC/C++JAJavaOOOOPsDSDSAOPOperating SystemsCOComputer Networks

Core Computer Science Subjects

AI / ML

MAMachine LearningSUSupervised LearningUNUnsupervised LearningDEDeep LearningANANNCNCNNRNRNNNLNLPAGAgentic AILLLLMRARAGPRPrompt EngineeringFIFine-tuningMOModel Evaluation

Training, Modeling, language systems, agents, retrieval, and evaluation.

Data Engineering

ETETL/ELTDAData ModelingFEFeature EngineeringDAData CleaningDAData Validation

Data preparation, validation, feature design, and pipeline thinking.

MLOps & Deployment

FAFastAPIDODockerAWAWSAPAPI IntegrationMOModel DeploymentMOModel VersioningMOMonitoring BasicsGIGitLILinuxJUJupyter

Interfaces, packaging, monitoring basics, and deployment workflows.

Frameworks & Libraries

LALangGraphHUHuggingFacePYPyTorchTETensorFlowKEKerasSCScikit-learnPAPandasNUNumPyOPOpenCVMAMatplotlibSESeabornSTStreamlit

Practical libraries used in building AI/ML Applications.

About

Hello and Welcome

Thanks for visiting my portfolio. We are living in one of the most interesting times in human history. AI is becoming a part of our everyday lives and is changing the way we learn, work, and create. The future ahead will be very different, and I believe AI will play a major role in shaping it. Human curiosity has always pushed us to build new things and solve difficult problems. That same curiosity is what drives me. I want to contribute to building this future and be part of the people creating the next generation of AI. I enjoy learning how AI works, exploring new ideas, and building projects that help me understand this field better. There is still so much to discover, and I am excited to keep learning and growing with it.

Location

Nashik, Maharashtra

Focus

Applied Machine Learning, LLMs, Transformers, Agentic AI

Working Style

Start Building, do research build again keep repeating.

Languages

English, German, Hindi, Marathi

Education

Vellore Institute of Technology, Vellore

2025 - 2027

M.Tech Computer Science and Engineering (Artificial Intelligence and Machine Learning)

Currently pursuing

K.K. Wagh Institute of Engineering and Research, Nashik

2021 - 2025

B.E Artificial Intelligence and Data Science

Completed

Contact

Connect with me, we can learn, explore and build together.

Use any direct link below to reach me or review my work

Emailsankalp1borse@gmail.com GitHubgithub.com/SankalpBorse LinkedInlinkedin.com/in/sankalp-borse LeetCodeleetcode.com/u/Sankalp_