AI Researcher & Engineer

Adit Whorra

I'm an AI researcher and engineer building systems that create real-world impact. I have worked across climate modeling, animal welfare, and legal tech, all connected by one goal: applying AI to deliver tangible, positive change.

I'm currently at Wadhwani AI, India's largest AI for Social Good research organisation. I work on ASR models for Indic languages and contribute to tools that measure reading fluency for children at scale across India, helping millions become better readers. I hold an MSc in AI from TU Delft (Netherlands), supervised by Dr. Elvin Isufi, and a BSc from Ashoka University (India).

Beyond research, I build things to solve problems I see around me and write about the process. The gap between ideas and execution is shrinking fast and that changes what any of us can do about the problems we care about.

Outside of AI, I care about personal finance, building good habits, and mental and physical well-being.

Interests

AI for Social ImpactASR & SpeechAI SafetyBuilding with AIAI for Productivity

Scroll

Latest News

Recent milestones, publications, and career updates.

February 2026
Represented Wadhwani AI at the India AI Summit
Participated in one of India's flagship AI policy and innovation gatherings, representing Wadhwani and engaging with government, industry, and research stakeholders.
February 2026
Attended the Bharat Bodhan Conclave
Represented Wadhwani AI at the conclave organised by IIT Madras, connecting with emerging startups and organisations working at the intersection of AI and education across India.
January 2026
Visited Ashoka University for AI for Social Impact Networking
Engaged with researchers and students at Ashoka University exploring how AI can drive large-scale social impact in India, sharing perspectives from ongoing work at Wadhwani AI.

Latest Writing

Thoughts on AI, research, and building things that matter. These are older posts, but more are on the way.

AI SafetyAI AlignmentScalable Oversight

An Intuitive Introduction to Solving Scalable Oversight using Iterated Amplification

February 27, 20245 min read

How do you supervise an AI that's smarter than you? Iterated Amplification offers a surprisingly elegant answer — break the problem down, align the pieces, and build up from there.

NLPAdversarial TrainingDeep Learning

Everything You Need to Know About Adversarial Training in NLP

January 4, 202113 min read

Adversarial examples expose fundamental limitations of deep neural networks. This post covers what they are, how adversarial training works, and why the robustness-generalisation trade-off is harder to solve than it looks.

Read All Posts

On My Shelf

Books I've read, am reading, and want to read.

Machine Learning System Design Interview

Human Compatible

Uncontrollable

Technopoly

AI Ethics

Atomic Habits

Mindset

The 7 Habits of Highly Effective People

The Comfort Book

How to Calm Your Mind

The 5 AM Club

The Psychology of Money

Same as Ever

The Richest Man in Babylon

The Intelligent Investor

The Power of Now

The Emergent Mind

The Anatomy of Hate

Age of Unpeace

The Tattooist of Auschwitz

Chaos

A Brief History of Time

Featured Projects

Selected work spanning research, industry, and open-source.

Climate PolicyMulti-Agent RLAI for Social Impact

Climate Policy Equity Framework

Climate negotiations fail partly because models used to guide them optimise for a single objective and ignore trade-offs between growth, temperature, and fairness. We built JUSTICE, a framework that plugs multi-agent, multi-objective RL into an Integrated Assessment Model so you can actually see those trade-offs. The result is a richer set of policy options that puts equity on the table alongside the usual economic and climate targets. Published at IJCAI 2025.

Paper Code

Spatio-Temporal ForecastingGraph ML

Spatio-Temporal Simplicial Complexes for Multivariate Time Series

My MSc thesis. Standard graph neural networks only model pairwise relationships, which misses a lot of what's actually happening in real networks like traffic or energy grids. I extended GNNs to higher-order topologies using simplicial complexes, letting the model capture interactions between groups of nodes, not just pairs. Benchmarked on traffic, energy, and air pollution data against 10+ baselines and got competitive results with a much leaner model. Awarded 9.0/10, supervised by Dr. Elvin Isufi.

Thesis Code

Weather ForecastingGraph ML

GraphCast Lite: Scalable GNNs for Weather Forecasting

DeepMind's GraphCast is impressive but expensive to run. This project looked at whether you can make it leaner without hurting forecast quality. We introduced a dynamic pruning approach that drops low-value edges during training, which cuts compute noticeably while keeping accuracy almost identical. Tested on large-scale meteorological data.

Code

Formal VerificationReinforcement LearningAI Safety

Formally Verifying Decomposed Reinforcement Learning

RL agents are hard to trust in safety-critical settings because you can't easily reason about what they'll do. We broke complex tasks into smaller sub-policies and formally verified each one using finite-horizon state distributions. You get provable safety guarantees per sub-task without sacrificing too much performance. Tested on Minigrid. A small step toward RL you can actually deploy in the real world.

Code

View All Projects