Pipeline · NLP
ClaimWatch
Claim detection over public message streams: extracts factual claims, scores them by check-worthiness and reach, and routes only the top of the queue to human reviewers.
Nate · AI/ML · Agentic Systems
I build large scale AI systems
Truly, I love building high-throughput data platforms and the AI systems that run on top of them: pipelines measured in millions of records a day, models placed where they earn their cost, and agents that perform actions safely and can not quietly disable their own oversight.
/about
Scale is a design constraint, not a headline.
My work sits where data platforms meet applied ML: ingesting large volumes of messy, heterogeneous records, normalizing them into something queryable, and layering extraction, classification, and retrieval on top. At small volumes almost any approach works. The interesting engineering starts at the point where the naive version gets too slow, too expensive, or too wrong to trust.
The problem I keep coming back to is entity resolution — deciding that two records from systems that were never designed to agree describe the same thing. It's probabilistic, it resists clean evaluation, and being confidently wrong is far more costly than being uncertain.
I also run a small agent stack on my own hardware: local models, scheduled jobs, and oversight built as infrastructure rather than instruction. Alongside that I build detection pipelines for public data and make generative art when I want a problem that ends the same day I start it. Python is where I'm fastest.
Tier the work
Lightweight models handle the bulk. LLMs are reserved for what's already been flagged. Cost and latency are architecture decisions, not line items you apologize for later.
Guardrails in code
Enforced through credentials, token scopes, and OS-level controls. A prompt is a suggestion, not a control.
Separate the oversight
On my own agent stack, monitoring and audit logs run on physically different hardware than the agent. Nothing should be able to switch off its own supervision.
/resume
Experience
Download PDF ↓Feb 2023 — Present
Applied ML Associate, Senior
JPMorgan Chase & Co. — Global Private Bank AI/ML · Jersey City, NJ
- Built an agentic LLM system that labels and categorizes electronic communications for legal e-discovery, cutting manual review overhead.
- Led a knowledge-graph ML pipeline end to end on TigerGraph — from connectivity infrastructure (PrivateLink, EC2, NLB, mutual-TLS) to a graph-based LSTM with multi-head attention deployed on EKS.
- Re-engineered a serverless inference platform processing 3–4M messages/day across 15 supervision policies, cutting batch time 70% (EventBridge, ECS, Lambda, RDS, S3, EKS).
- Shipped a RAG review-assistant chatbot on a MongoDB vector store and set the team's MLOps deployment standard (FastAPI, Pydantic, Docker, Kubernetes, Jenkins).
Aug 2019 — Jan 2023
Software Engineer
Carnegie Mellon University, Software Engineering Institute · Pittsburgh, PA
- Lead engineer on A3ITE — standardized assembly, deployment, and monitoring of data-processing pipelines (Python, VueJS, Docker, Kafka, NiFi).
- Built the AI End-to-End platform for secure, generalized ML deployment on an air-gapped network (Flask, NiFi, Neo4j, Prometheus, Grafana).
- Earlier, as data engineer, built an ML pipeline studying social-media spambot influence on trending-topic sentiment.
Selected work
Infrastructure · Local LLM
Local agent stack
A morning briefing agent running on a dedicated mini PC via Ollama, delivered over a Tailscale mesh — with guardrails, monitoring, and audit logs deliberately hosted on separate hardware.
Product · Micro-SaaS
westforge.io
An early-stage B2B engine that turns podcast appearances into qualified pipeline. Data model designed across 13 entities; currently in concept and validation.
Generative · p5.js / py5
Between Tokens
Programmatic video and generative art experiments — the side of the practice where the output is meant to be looked at rather than queried.
Toolkit
Python
SQL
Terraform
AWS
Docker / Kubernetes
LangChain
RAG / Agentic
TigerGraph / Neo4j
MongoDB
Kafka
FastAPI
Credentials
DoD Top Secret / SCI clearance · SEI Software Architecture Professional · AWS Cloud Practitioner
Education
May 2019
B.S. Computer Science
Virginia Commonwealth University · Richmond, VA — Minor in Mathematics
/contact
Let's talk.
Open to conversations about platform engineering, applied ML, and agentic systems — especially the kind that have to survive contact with real volume.
west.nrh@gmail.com