I build AI systems that decide what to do next.

Case files.

• 01
  Production · GPU service

  Turning long, ugly PDFs into structured data

  An asynchronous extraction service for a maritime software platform. Long technical documents go in; structured blocks, markdown and search-ready chunks come out. The customer's file is read once and never stored.

  TypeScriptPythonSQS · S3 · EC2PostgreSQL↗
• 02
  Production · Agent

  An agent that cannot leak another tenant's data

  Natural-language search over a graph of candidates for a specialist staffing platform. One tool-calling loop, and between the model and the database sits a deterministic gate.

  TypeScriptLangGraphNeo4jFastify↗
• 03
  Production · Modelling

  Forecasts that admit how wrong they can be

  A fuel-consumption model for one vessel class, served through a small web console. The work that mattered most was the audit that came after the first version.

  PythonFastAPIscikit-learnConformal prediction↗

Three things I won’t compromise on.

• Confidence is not verification.

  A model will rate a wrong answer as certain. I check outputs against something outside the model.

• Refusal is an outcome.

  A system that can't say “I can't support that” will eventually say something false.

• Measure it, then print the misses.

  Ablations, held-out sets and failure audits belong in the write-up, not in a drawer.

Where things stand.

Now

Owning production AI systems end to end at a maritime software company, as an engineering intern with a senior's scope.

Studying

B.Tech in Computer Science (AI & Data Science) at IIIT Kottayam, 2023–2027. IIT Madras BS, foundation level completed.

Earlier

PromptForge, an agentic prompt-to-system platform. Paused. What I learned there is in everything I've built since.

Quietly

On the side, helping very old books find new readers.