Suresh Raghu
Independent ML Researcher · AI Lead at VFS Global
Open to PhD · Fall 2027
New Delhi, India
I’m an independent researcher working on uncertainty quantification and reasoning reliability, when and why large language and vision-language models fail to know what they don’t know. My work spans reasoning LLMs, multimodal models, and agents, with publications accepted at the ICML 2026 workshops on Failure Modes in Agentic AI (FAGEN) and Combining Theory and Benchmarks (CTB).
Separately, by day I’m the lead AI engineer at VFS Global, where I build production AI systems that process over a million documents daily across 110+ countries.
I hold a B.Sc. in Programming & Data Science from IIT Madras along with a B.Tech in Computer Science and Engineering from VIT Bhopal. Always happy to talk research or engineering, feel free to reach out.
news
| May 25, 2026 | Two papers accepted at the Failure Modes in Agentic AI (FAGEN) workshop at ICML 2026: SELFDOUBT and Proper Scoring Rules for Agentic Uncertainty Quantification! |
|---|---|
| May 22, 2026 | Proper Scoring Rules for Agentic Uncertainty Quantification was accepted as a poster at the Combining Theory and Benchmarks (CTB) workshop at ICML 2026. See y’all in Seoul! |
| May 06, 2026 | New preprint: Proper Scoring Rules for Agentic Uncertainty Quantification. We introduce the Trajectory Proper Score (TPS), a predictor-agnostic family of strictly proper, trajectory-level scoring rules for evaluating uncertainty in LLM agents. |
| Apr 07, 2026 | New preprint: SELFDOUBT: Uncertainty Quantification for Reasoning LLMs via the Hedge-to-Verify Ratio. We introduce the Hedge-to-Verify Ratio (HVR), a single-pass uncertainty signal for reasoning LLMs that outperforms Semantic Entropy at about 10x lower inference cost. |
| Apr 05, 2026 | New preprint: Don’t Blink: Evidence Collapse during Multimodal Reasoning. We identify evidence collapse, a decay of visual grounding during multimodal reasoning that text-only uncertainty signals cannot detect. |
| Jan 11, 2026 | Published an article on prompt ordering in vision-language models: Why Your VLM Prompts Are Backwards (And How to Fix It) |
| May 02, 2024 | Joined AI @ VFS as lead AI engineer |
latest posts
selected publications
-
2026Accepted as a poster at the CTB and FAGEN Workshops at ICML 2026 -
2026Accepted as a poster at the FAGEN Workshop at ICML 2026. Shows better uncertainty quantification performance than Semantic Entropy at about 10x lower cost
