Finalist of a 30-hour offline hackathon held at Thadomal Shahani College Of Engineering, Bandra, Mumbai.

Participated in Xylem Ignite International Hackathon,collaborating with diverse teams to solve real global challenges.

Participated in Devignite, Navi Mumbai's first offline hackathon organized by Smt. Indira Gandhi College of Engineering.

Secured 5th prize at Anveshna Mumbai 2023, sponsored by Synopsys.

ElderConnect was selected as a finalist project at HACKX 2023, organized by S.F. Institute of Technology.

Won the international project competition at PICT, Pune (Impetus and Concepts) with trophy, certificate, cash prize, and vouchers.

First prize in the project competition at Loknete Gopinath Munde College of Engineering, Nashik.

Participated in the divisional level project competition organized by IEEE Bombay at Godavari College of Engineering, Jalgaon.

Real-time ASL recognition system using VideoMAE, PyTorch and VideoAug on a 20GB dataset with 239 signs. Achieved ~98% classification accuracy and built a multi-sign inference pipeline for long-form videos.

Implemented a GPT-style LLM with custom BPE tokenization, multi-head attention, sampling (temperature, top-k, top-p), gradient clipping and a full training pipeline.

Built a self-driving car model with behavioral cloning, road sign detection and real-time decision-making using CNNs, TensorFlow/Keras, Flask, Socket.IO and Udacity’s open-source simulator.

End-to-end classification pipeline to predict mobile price category using EDA, preprocessing, model training and a SageMaker-ready script for cloud training and deployment.

Cold email automation tool using Groq, LangChain and Streamlit. Scrapes a company’s careers page, extracts job listings and generates personalized cold emails with portfolio links from a vector DB.

Platform to connect elders with volunteers via one-to-one and group video calls, emotion recognition with AI, and event scheduling to reduce loneliness and improve engagement for elders.

Production-style abstractive text summarization system using Hugging Face Transformers (T5/BART), with modular YAML configs, logging, and a stage-wise NLP pipeline.

Full MLOps pipeline for phishing / network threat detection with data ingestion, validation, transformation, training, MLflow tracking, batch prediction, Docker, CI/CD (GitHub Actions) and AWS (S3 + EC2) deployment.