AI systems, agent harness, and the data that feeds them.

I build the infrastructure that makes a model good at a real task: agent harnesses, MCP servers, and pipelines that turn crawlers and APIs into data the agent can act on.

View my CV for details and rates.

ls ./services

01 agents
AI Agent Systems

Agent harness infrastructure: wrapping LLMs with tools, APIs, and memory so they can browse, act, and run multi-step workflows autonomously.
02 mcp
MCP Servers

Design and ship Model Context Protocol servers that expose your data and tools to agents safely - read-only query surfaces, guardrails, streamable HTTP.
03 crawl
Web Scraping at Scale

Distributed Rust crawlers fetching millions of pages and API endpoints daily, with a small and predictable infrastructure footprint.
04 pipelines
Data Pipelines · Rust + Python

Ingestion → processing → delivery pipelines: Rust for high-throughput, low-footprint stages, Python for flexible processing and glue.
05 single-view
Single-View Data Systems

Consolidating many sources into one clean, structured, queryable source of truth - breaking down silos for analytics and decisions.
06 datasets
ML / LLM Dataset Preparation

Collecting, cleaning, and delivering high-quality datasets for training and evaluating specialized language models and AI systems.

cat approach.md

From acquiring information at scale to processing it to the analytics on top, I've worked the full data lifecycle - that end-to-end command is what makes the data actually useful.

Rust where it counts. Crawlers, queues, parsers, plus the CLI tools and MCP servers agents call directly - all in Rust for speed, low footprint, and reliability. Python for flexible processing and glue.
A harness, not just a model. The harness is the infrastructure around a model: tools, APIs, memory, MCP surfaces - that makes it good at one specific job.
Built to run at scale. Distributed systems pulling millions of pages and API endpoints a day on a small footprint, with pipelines that keep the data consistent downstream.

ls ./open-source

Terminal screensavers in Rust

tarts

Blazingly fast, memory-safe terminal screensavers - Matrix rain, boids, plasma, and more.

CETD content extraction

dom-content-extraction

Extract main content from HTML via text density - Unicode-aware, with markdown output.

Exoplanet catalog + MCP

exodata

NASA exoplanet catalog (Leptos, Axum, Polars) with a REST API and a hosted MCP endpoint.

Async framework for crawlers

capp-rs

Tower-native async runtime for web crawlers and task pipelines - queues, retries, backpressure, DLQ.

Page analysis for AI agents

pageinfo-rs

HTTP-only page analysis for LLMs and agents - identity, URL structure, metadata, feeds, embedded JSON.

Probabilistic data structures

probabilistic-rs

Bloom and expiring-Bloom filters with Python bindings, REST API, and CLI/TUI - ~15-17M ops/sec.

AI systems, agent harness, and the data that feeds them.

AI Agent Systems

MCP Servers

Web Scraping at Scale

Data Pipelines · Rust + Python

Single-View Data Systems

ML / LLM Dataset Preparation