Founder & CTO, Craton Labs · verifiable AI for banks, insurers and fintechs
Now going deep on AI sandboxing
AI sandboxing
Containers and verifiers that coding agents can't escape or game.
RL environment workbook →
RLHF & post-training
PPO, GRPO and DPO, and verifiable rewards that resist hacking.
Frontier AI study tracker →
AI optimization
GPU kernels, memory math and efficient serving of large models.
Frontier AI study tracker →
Verifiable AI and data systems for regulated finance.
AI engineering
An extraction pipeline for regulated documents where every value in the output grid carries a citation back to the document that produced it — and a verification layer that decides whether a value is allowed to stand
AI engineering
Claim-level grounding, unsupported-claim rate, and adversarial testing — the three measurements that tell you whether an LLM output can be relied on, rather than merely sounds right
projects
A deep dive into e-commerce data using K-Means clustering to uncover the hidden relationship between delivery performance and customer retention.
From data pipelines to verifiable AI.
2021 - 2023
Siemens Ltd.
ETL • Python • Data Pipelines
Jul 2024 - Dec 2024
Habitat
Edge Inference • CI/CD
Jan 2025 - Sept 2025
RiskPe
PySpark • SQL • Power BI
Oct 2025 - Present
Craton Labs
BFSI AI Compliance • RAG Evaluation
ML and statistics, implemented from scratch.
Statistics
Building a complete experimentation framework — from hypothesis testing to Bayesian inference
ML
Understanding the mathematics of reducing high-dimensional data — from eigendecomposition to manifold learning
ML
Implementing core ML algorithms from first principles using only NumPy