I build AI for situations where being wrong is expensive — a mule account that clears,
a contract clause extracted incorrectly, a statistic published to a nation. That constraint shapes
everything: how the system retrieves, how it explains itself, where a human signs off.
Distributed systems
→Financial crime AI
→Banking-scale ML
→GenAI & AI architecture
I started at Knoldus writing Scala, Kafka and Akka services alongside ML
work — which is why production concerns have never felt like someone else's job. At
Tookitaki I specialized: eighteen months of AML modeling for banks and payment providers
across APAC, Japan and Europe, covering transaction monitoring, name screening, customer risk and
network analytics. At Barclays I applied that at banking scale, inside a global bank's
model governance environment, on mule detection and transaction monitoring.
At Presight.ai the work widened into GenAI and architecture: agentic systems
over national statistical data, document intelligence across regulatory corpora, knowledge graphs, entity
risk scoring, and sovereign LLM deployment into air-gapped environments. I was promoted to Senior Data
Scientist in April 2024, and increasingly own architecture-level decisions rather than individual models.
The through-line is a preference for problems where accuracy, explainability,
security, regulation and production reliability all have to hold at once. Those systems are
slower to build and considerably more interesting.
