Matchy Matchy is an e-commerce discovery platform. Our interfaces today are a conversational agent aura & a traditional search engine. Matchy Matchy is an ambitious deep tech product in the search space; building a fundamentally new type of search.
Obsess is an experiential e-commerce platform. The company was recently acquired for $25M by Infinite Reality. I worked as a Senior Product Manager on their e-commerce platform. I lead the 0-1 development of the platform, which played a key role in the acquisition.
A fully local knowledgebase that automatically extracts and indexes content from every page you browse. The system uses Embedding Gemma for semantic embeddings and implements a hybrid search architecture for retrieval. The RAG interface is powered by Chrome's built-in Gemini Nano, keeping everything on-device and private.
An end-to-end video analysis platform designed for large-scale video content. Built on Elasticsearch with a multimodal embedding model, it enables semantic search and retrieval across video content. Production-ready and fully deployable.
A non profit project built to provide access to frontier LLM models to underprivileged communities in India. Has deep research and RAG capabilities.
Mydocc was a primarily academic project I worked on while at Grad school. We used a computer vision model to digitize healthcare records in India. The project ultimately ran into data compliance issues and was never fully commercialized/launched.
I however, learnt how to do product things- talk to users, work with engineers, design the user experience and design the system architecture.
A free, open-source tool that scrapes product data from e-commerce sites and upserts it into a vector database. Built with crawl4ai, it provides the foundation for building advanced search solutions and RAG agents. The tool is adaptable to different data structures, making it versatile for various applications beyond e-commerce.
A system based on the recent Anthropic paper on the same topic that improves accuracy of RAG agents. Adds "context" to every piece of chunked data for more accurate retrieval. Built the functionality to ingest a piece of data, chunk the said data, add context to each of the chunks and embeds it into a vector a database.
A simple dev tool that allows developers to get the markdown of a webpage. Very helpful when working with LLMs particularly for quickly copying documentation.
An experimental art website exploring the vulnerability felt when unsending a typed message.