GAIS
Control your computer with only your eyes
Madhavendra Thakur
fellow @ y combinator
cs + econ @ harvard
published ai research @ neurips, iclr, icml
01
Projects
Proficiencies
React
Next
Electron
JavaScript
HTML
CSS
Tailwind
Python
PyTorch
TensorFlow
500+ daily active users
TrinHub
Comprehensive enterprise educational content management system
1000+ phrases
Vaach Languages
AI-powered translation software for low-resource languages
Forj
Create bespoke datasets on demand
3.3k+ installs
Writeboard
A whiteboard in your browser
Presented @ MIT
macrocycle_design
AI to design drugs for "undruggable" cancers
Sticker Tag
An online platform for a schoolwide game.
Published on PyPI
liver_annotation
Python library to parse liver cell data with ease.
02
Experience
Full Stack Software Engineer, Systems Architect, UI/UX Designer
Built software platforms for commercial clients, including a Gemini integration layer for ElasticSearch, ML engineering benchmarks for METR, and an end-to-end tournament management system for the world's largest quiz bowl tournament operator.
Research Intern
Developed ML pipelines to design peptide drugs for traditionally undruggable cancers, working alongside biochemists toward manufacturable therapeutics. Presented the work at IEEE MIT URTC 2025.
03
Research
Reasoning Isn't Enough: Examining Truth-Bias and Sycophancy in LLMs
Collaboration with Columbia AI Alignment. Analyzing the truth-bias of reasoning LLMs across socioeconomic contexts
Towards Data Governance of Frontier AI Models
Collaboration with Center for Human-Centric AI. Proposes data governance as a practical lever for shaping the development and deployment of frontier AI systems.
Opportunities and Challenges of Synthetic Data Governance
Examines the viability of governing synthetic data as a proxy for frontier model development.
Towards Neural No-Resource Language Translation: A Comparative Evaluation of Approaches
Compares approaches for translation in "no-resource settings", focusing on performance tradeoffs when conventional reference data is unavailable.
Governing Automated Strategic Intelligence
Collaboration with Center for AI Safety. Examines how frontier AI enhances OSINT, "AUTOINT", and how governments should approach it.
Culturally-Grounded Chain-of-Thought (CG-CoT): Enhancing LLM Performance on Culturally-Specific Tasks in Low-Resource Languages
Collaboration with Stanford NLP. Introduces culturally grounded chain-of-thought prompting to improve LLM performance in low-resource language settings.
Single Cell Transcriptomics Reveal Genomic Indicators of Nonalcoholic Fatty Liver Disease
Uses single-cell transcriptomics to identify genomic signals associated with nonalcoholic fatty liver disease.