I build the pipelines behindthe boxes that move the world.— and the ML systems that make sense of them.

Tripwire

Real-Time LLM Output Supervisor · Streaming Guardrail

Date

OSS · MIT

Role

Author

A real-time supervisor for streaming LLM output: it watches tokens as they generate and aborts a harmful response mid-flight — killing a bad 3,000-token answer at token ~200 instead of paying to generate the whole thing and rejecting it after.

Built the streaming supervisor and a one-call Guard API — shipped to PyPI as tripwire-llm — with TypeSafe's Jev, a ~150 ms System One model, as the in-path detector. Per-category safety thresholds across ten signal types, plus a live, access-gated demo. The calibrated judge is fast enough to sit in the token path, where a second LLM acting as judge can't keep up.

• Python
• FastAPI
• asyncio
• TypeSafe Jev
• Next.js
• Vercel

~150ms

In-path judgment

10

Safety signal types

PyPI

Published package

SARA

Multi-Agent RAG · ServiceNow Incident Resolution

Date

Jun 2026

Role

Agent orchestration · retrieval

A production multi-agent RAG pipeline that classifies and root-causes ServiceNow incidents, then routes each to the right engineer.

Agent orchestration plus a hybrid retrieval path — Qdrant vector search with Cohere reranking — reading two knowledge collections across four incident categories.

• LangGraph
• Azure OpenAI
• Qdrant
• Cohere Rerank
• FastAPI
• Docker

−75%

Mean resolution time

94%

Classification accuracy

300+

Engineer-hrs / month saved

RateFlow

Cross-Platform API Gateway · Terminal Operations

Date

Apr 2026

Role

API design · data sync

A governed data-access layer exposing terminal and rate data to internal platforms through versioned API contracts.

Three endpoints — TMVC, PO Invoice, Container Info — with schema validation across CDT/PP/Prod, plus an event-driven Postgres → Delta Lake rate sync that picks up changes in real time.

• Python
• FastAPI
• OpenAPI 3.0
• StarGate / Apigee
• PostgreSQL
• Delta Lake

50K+

Requests / day

10K+

Rate updates / day

−20 hrs

Manual work / week

Delta Lake Platform

Container Analytics · Databricks Lakehouse

Date

2025–26

Role

Data engineering

The fleet of 30+ PySpark pipelines on Databricks Delta Lake that feeds real-time freight analytics for container operations.

Ingestion and dimensional models that carry shipping records off every terminal into the lakehouse. Reworking the ingestion path cut end-to-end latency and steadied the analytics reading downstream.

• PySpark
• Databricks
• Delta Lake
• Azure Data Lake
• Dimensional Modeling

12M+

Records / day

30+

Global terminals

−70%

Data latency

researgent

Multi-Agent Research System · CLI

Date

OSS

Role

Author

A CLI research assistant that plans, searches the web, and synthesizes cited answers through multiple specialized agents.

Configurable agent count and research depth, four output styles (academic / business / educational / journalistic), source attribution with citations, JSON export, and batch + interactive modes.

• LangGraph
• LangChain
• Google Gemini
• Tavily
• aiohttp

4

Output styles

N-agent

Configurable depth

TELEDOC

AI Symptom-Diagnosis Web App

Date

OSS · MIT

Role

Author

A web app that turns symptom selection into AI diagnostic suggestions and medication recommendations.

A React front end over a FastAPI + LangChain service that structures model output into ranked, explainable suggestions.

• React
• styled-components
• FastAPI
• LangChain
• Google Gemini

MIT

Open source