Selected work / 2022—now

Built, tested, and put to use.

Projects in applied machine learning and agent systems, shaped by the parts that have to work in practice.

Agentic Commerce

Sellable

Sellable is an agentic commerce prototype where buyer and seller agents negotiate within explicit policy limits. It combines a machine-readable storefront, deterministic transaction checks, single-use consent, Razorpay test-mode payments, and an audit trail.

ML Visualization

Confluence

Case study

Confluence is an interactive machine-learning playground powered by real scikit-learn compute. It pairs 38 algorithms and 25 datasets with live decision boundaries, training playback, prediction explanations, and reproducible Python exports.

Agent Harness

Helion

In progress

Helion is an agent harness in development for terminal and desktop workflows. It is designed to show agent runs, allow mid-run steering, and gate tool calls for human approval.