I’m Swapnil — an AI Scientist at Intuit, where I work on multi-agentic LLM systems, retrieval-augmented generation, and the alignment problems that come with deploying large models in high-stakes financial contexts. Before Intuit I was at IBM Research on the core team for watsonx.ai’s inference engine, and before that I did an MS at NYU. I shipped the first MCP implementation at Intuit as part of the Intuit x Anthropic partnership, and I occasionally surface in Kaggle competitions (Expert).
My research is on mechanistic interpretability, adversarial robustness, and agentic systems — understanding why models behave the way they do, and what it takes to break them.
Resume · Google Scholar · GitHub · LinkedIn · Email
Research Interests
- Mechanistic Interpretability — Circuit-level analysis of transformer models, understanding which components encode which behaviors, and making that analysis reproducible and uncertainty-aware (CIRCUS).
- Adversarial Robustness — Universal adversarial triggers, attack transferability under model compression, and multimodal safety. I’m interested in the gap between empirical robustness claims and real-world deployment.
- Agentic and Latent Reasoning Systems — Backdoor attacks on chain-of-thought and latent reasoning models, steering and safety for systems that reason in ways we can’t directly observe.
On Music
“Simplicity is the final achievement. After one has played a vast quantity of notes and more notes, it is simplicity that emerges as the crowning reward of art.” — Frédéric Chopin
| Chopin — Ballade No. 1 in G minor, Op. 23 | Chopin — Nocturne in E-flat major, Op. 9 No. 2 |
| Bach — The Well-Tempered Clavier, Book I | Bach — Goldberg Variations, BWV 988 |
| Beethoven — Piano Sonata No. 14 “Moonlight”, Op. 27 No. 2 | Beethoven — Piano Sonata No. 23 “Appassionata”, Op. 57 |
| Ravel — Pavane pour une infante défunte | Ravel — Gaspard de la nuit |
| Tchaikovsky — Piano Concerto No. 1 in B-flat minor, Op. 23 | Schubert — Impromptu in G-flat major, Op. 90 No. 3 |