Hi there, I'm Tanmay!!

Sep 2024: Our multi-turn conversation attack paper is now arxiv

RED QUEEN: Safeguarding Large Language Models against Concealed Multi-Turn Jailbreaking

May 2024: Joined HippocraticAI as an Applied Scientist working on Speech LLMs 🚀🚀🚀

March 2024: Built Subtext that won in the Chat Interface category at AGI House Hackathon

Introducing Subtext! 🚀📱 Our chat interface redefines texting with AI-generated text animations and emojis on the fly! 🤖🎉 How do we do it? 🤔 We prompted Gemini to format incoming text with HTML, CSS, and JavaScript animations, then rendering it in a browser on the fly! 🌐💡

But here's the real kicker - Subtext goes beyond the cool factor! 🌈 Our creation focuses on accessibility, catering to individuals with ASD or similar challenges, ensuring everyone can grasp the meaning and emotion behind texts, even humor and sarcasm (remember Sheldon from Big Bang Theory?) ! 🧩❤️

LinkedIn Post

January 2024: Won 2 Hackathons!

suRAGeon - AI Assisted Medical Surgery (Note: code was built in a day for a hackathon, so its a pretty hacky solution)

X-Ray Insight - Evaluating Multimodal RAG Pipeline for question answering on XRay Images

The RecWizard libary is built for easy deployment/development of Conversational Recommender Systems(CRS). The core idea of RecWizard is to use a "Text-in, Text-out" interface for all kinds of different modules used in a CRS pipeline. The text-based interface gives our package a plug-and-play nature that allows users to easily experiment with new combinations of modules. Built with awesome folks at McAuley Lab UCSD.

May 2023: Joined Megagon Labs as a Research Engineer

December 2022: ClimaBench work accepted at NLP4PositiveImpact workshop at EMNLP 2022

The topic of Climate Change (CC) has received limited attention in NLP despite its real-world urgency. Activists and policy-makers need NLP tools in order to effectively process the vast and rapidly growing textual data produced on CC. Their utility, however, primarily depends on whether the current state-of-the-art models can generalize across various tasks in the CC domain. In order to address this gap, we introduce Climate Change Benchmark (#ClimaBench), a collection of datasets for evaluating model performance across a diverse set of CC NLU tasks systematically.

September 2022: New Ranking Explainers in OmniXAI

January 2021: NLP For Climate Change

Led a team of 50 international collaborators that developed Heatmaps (hazards v/s solutions), Q&A system and Knowledge Graphs to assess coalition platforms on climate adaptation, and match key climate risks to adaptation measures

• Q&A – Implemented novel search engine on Dense Passage Retrieval (DPR) finetuned embeddings of documents and indexed using Elastic Search deployed on AWS EC2. DPR outperformed existing BM25 based indexing (0.87 v/s 0.74 MRR on retrieved docs). Read more

January 2021: Climate Change Actors Network Analysis

October 2021: AI For Malaria Prevention

Built an Ensemble model using features from landcover image segmentation U-Net model representations, population density, DeepWaterMap and topographical features generated from Google Earth Engine data for Ghana, Africa. Detected stagnant water bodies with 0.905 AUC. (in collaboration with Award winning startup Zzapp Malaria. Read more

COVID19 3D Globe Tracker

Let's Connect 😃