I build systems
that handle money,
trust, and identity

If you build with LangChain or the OpenAI Agents SDK, every prompt and every tool call routes through a US server. For teams that need data sovereignty — regulated industries, anyone billing in a non-US currency — that's a non-starter.

A self-hostable runtime where every long-running workflow can pause durably for a human to approve, reject, or rewrite. Memory survives restarts. Sensitive actions produce receipts. Prompt injection is caught at the gate, not after the LLM has already acted on it.

Public on GitHub, MIT-licensed, currently v0.4.0. Built for teams that need their agents to use local payment rails — M-Pesa, IntaSend, whatever the market actually runs on — and keep their customers' data on infrastructure they own.

Developers were building AI agents with no distribution. Buyers had no trusted way to find, evaluate, and pay for agents that actually ran tasks — not just demos.

Two-sided marketplace: developers list agents, buyers hire and execute them through a managed proxy. Built-in trust engine with KYC, anomaly detection, and automated trust scoring. Payments in USDC. Browser never touches Django directly — BFF architecture throughout.

In active development. The first marketplace I know of that's purpose-built for agent-to-agent commerce — real payment rails, real identity verification, not just a directory of links.

A B2B distributor was managing bulk orders, agent commissions, and M-Pesa payments entirely through manual WhatsApp threads. No records, no reconciliation, no scale.

Full B2B platform with a multi-agent ordering system, product catalogue, M-Pesa STK Push payment collection, and order management dashboard.

Shipped and live. Replaced the manual process entirely — orders, payments, and agent tracking in one place.

Every Django developer building for the Kenyan market had to reverse-engineer the same M-Pesa Daraja 2.0 API from scratch. No clean library existed.

Published Python library that wraps the full Daraja 2.0 API for Django. STK Push, C2B, B2C, reversals — all abstracted behind clean Django REST patterns.

Used by other developers building payment-enabled apps in Kenya. The library you write because you got tired of writing the same M-Pesa boilerplate at the top of every new project.

Crypto traders had no automated way to detect coordinated manipulation. Pattern recognition was manual, slow, and always after the fact.

Analysis bot combining social-sentiment signals with on-chain market data to flag coordinated pump-and-dump patterns in near real-time.

Shipped, ran live through the 2022 altcoin season. The first project where machine learning, market-data pipelines, and security thinking all had to land together on the same codebase.