Projects
Community Contributions
Cloud Community Days Kolkata 2022 Website
Contributed to building an event website for the Google Developer Group Cloud Kolkata Community, aiding in attracting a substantial audience and making the event the largest in eastern India.
Valuable Contributors
Tech Stack :->
- TS/JS, React.js, TailwindCSS, Firebase, GitHub Pages
Winter of Code 2021 Website
Built an event website for Google DSC NSEC, titled ’Winter of Code,’ a month-long open-source event similar to GSoC.
Valuable Contributors
Tech Stack :->
- React.js, Material-UI, Bootstrap, GitHub Pages
Mentored Projects
Self Supervised Learning Study
Collaborated with an MLE in developing a streamlined pipeline to train model using self supervised algorithms. Used iNatualist-Aves dataset for the purpose. Followed proper modular coding style. Used WandB artifacts to store and visulaize dataset. Used WandB sweeps to optimize hyperparameters. Used TensorFlow similarity model SimCLR.
Mentored by Ayush Thakur
Valuable Contributors
Real Time Social Distance Detector
Built a real time social distance detector which is fast, accurate, and resource efficient, hence, can run on mobile devices. Used MobileDet model, an SSD (Single shot detector model) which is converted to TensforFlow Lite using TensorFlow Object Detection API for on-device inference. Used calibration (bird’s eye view) and euclidean distance to measure proper distance between pedestrians.
For more details: Check here.
Mentored by Sayak Paul & Rishiraj Acharya
Valuable Contributors
Competition
HungOver
Here is an idea behind this project. I along with teammates worked on this project in Google 2021 Solution Challenge.
Valuable Contributors
Personal
Quantized-LLMs
Applied Post-Training Quantization techniques on LLM models. Compared quantized models with non-quantized models by benchmarking them. Ran inference with quantized models on mobile CPU (iOS) and benchmarked them.
Tech Stack :->
- LLM, Auto-GPTQ, llama.cpp, lm-evaluation-harness, bigcode-evaluation-harness, swift-ui
PDFChatter
A locally running PDF chatbot where users can upload their documentation in pdf format and chat about it. It’s a Next.js app that read the content of an uploaded PDF, chunks it, adds it to a vector store, and performs RAG, all client side. You can even turn off your WiFi after the site loads.
Tech Stack :->
- LLM, Ollama, Langchain.js, Transformers.js, Chroma, Docker, Next.js, TailwindCSS, Vercel
Handlanmark detection
Here is an idea behind the project.
Tech Stack :->
- Mediapipe, React.js
Image Classification
Here is an idea behind this project.
Tech Stack :->
- TensorFlow.js, Node.js