AI Agent Development
LangGraph, Claude & Extraction
I help Utah and remote US teams ship AI that works in production: LangGraph agents, Claude-powered backends, document-extraction pipelines and RAG over private data—with evals, guardrails and cost controls.
LangGraph Agent Workflows
Production multi-step agents (LangGraph) that call tools, update systems and escalate when unsure—built for ops and field workflows, not open-ended chatbots.
- LangGraph state machines
- Tool definitions with least privilege
- Human approval gates
- Run logs for audit and debugging
Document Extraction Pipelines
OCR and LLM pipelines (LlamaCloud + Claude) that turn bills, invoices, POs and quotes into structured data with vendor matching and line-item extraction—anchoring real B2B spend-management products.
- LlamaCloud + Claude extraction
- Vendor matching
- Line-item schemas
- Human review queues for low-confidence rows
RAG & Claude-Powered Backends
Knowledge retrieval over your docs and databases, plus Claude-powered API backends with structured outputs, rate limits and clear UX for uncertainty.
- Hybrid keyword + vector search
- Citations and source highlighting
- Structured JSON outputs
- Cost and latency budgets
Quality, Safety & Ops
Evals against labeled examples, red-team checks for prompt injection and PII, monitoring for drift and spend so models do not quietly degrade.
- Offline + online eval harness
- Prompt-injection defenses
- PII handling guidelines
- Usage and cost dashboards
What You Get
- Working prototype or production feature against your data
- Architecture notes your team can extend
- Eval suite and failure playbook
- Optional MCP server if tools need a clean AI interface
- Handoff session with your engineers
How an Engagement Works
1. Scope the workflow
Pick one high-value use case, define success metrics and data sources.
2. Prototype with real data
Build a thin slice, measure quality and cost, decide go/no-go.
3. Harden for production
Add auth, logging, guardrails, evals and monitoring.
4. Hand off or retain
Document and train your team, or continue on a weekly retainer.
Frequently Asked Questions
What is RAG, and when do I need it?
+
Retrieval-Augmented Generation grounds an LLM in your documents, tickets or database instead of relying on training data alone. You need it when answers must cite your content, stay up to date, or avoid inventing company-specific facts.
How do you keep agents from taking unsafe actions?
+
Tool permissions are scoped, high-risk actions require human approval, and every run is logged. We add eval sets so regressions show up before production.
Can you add AI features to an existing product?
+
Yes. Most work embeds into current Rails, React or similar stacks rather than replacing them. We start with one high-value workflow and expand once quality and cost are under control.
Do you work with Utah companies on-site?
+
Discovery and architecture workshops can be in person around Salt Lake City. Day-to-day build work is usually remote with clear async updates.
Book a Free 30-Minute Consult
Tell me what you're building or what's stuck. I'll reply with next steps, usually within one business day.
Get In Touch
Why Choose Me?
- 10 years shipping production software
- 5 long-term fractional clients, most for 2+ years
- AI agents, MCP servers, Rails + React
- Based in Salt Lake City, working remotely US-wide