Subash Matu

Confer is an intelligent research assistant that simplifies academic papers by embedding LLM-powered helpers directly into the content, enabling intuitive understanding of complex concepts.

DJ Bruno is a real-time, fully-local personal DJ that curates music by understanding your emotional and situational context through multimodal sensing. It integrates facial expressions, voice input (like humming or singing), screen activity, and location cues to detect what you’re doing and how you’re feeling. These inputs power a central system that pulls from a vector database to build and continuously update a dynamic song queue. Bruno learns your preferences by observing patterns…such as when, where, and how you respond to certain songs, assigning emotional and contextual markers without the need for fine-tuning. A parallel process monitors for shifts in mood or activity, instantly refreshing the queue using song metadata. Users can also give feedback or make specific requests, allowing the system to adapt both in the moment and over time, creating a deeply personalized and responsive listening experience, all while keeping everything private and on-device.

Bruno Feels your Vibe before you do!

P.s Brown Bruno Bear be proud

Our project implements a deepfake detection model based on a paper that targets zero-day deepfakes, ones the system hasn’t seen before. Unlike traditional supervised classifiers, this approach detects inconsistencies between a deepfake and its associated claims (e.g., captions, identity, or audio-visual alignment) to flag manipulations, addressing the growing threat of misinformation and reputational harm.

An AI-powered tool that learns your unique speaking and writing style, suggesting words seamlessly. It continuously adapts, not just to your evolving language but also to the way you communicate with different people, just like you naturally do.

AutoPilot AI is a system that lets computers control themselves using PyAutoGUI and OmniParser under the hood. It automates tasks, navigates interfaces, and executes actions just like an AI-powered autopilot for your machine.

A multi-LLM ensemble framework designed to minimize hallucinations and boost factual accuracy by tapping into majority consensus across models. Built for reliability, this system enhances output trustworthiness through smarter model coordination.

CodePeek is a VS Code extension that generates a "Context File" listing your workspace structure and selected file contents and captures terminal output into a "TerminalContext File." It’s designed for sharing your code/project context with LLMs (like DeepSeekR1, ChatGPT, etc) or team members, making debugging, code reviews, and AI integration seamless. No complex setup…just click and create!