I build AI systems that ship and hold up in production.
Fractional work for teams with real workflows and real constraints. I take on a small number of engagements.
Start a conversationDocument search with citations
Answers from leases, specs, SOPs, contracts, legacy PDFs. Every response cites the exact clause.
Intake & extraction pipelines
Unstructured docs → structured data. Classification, extraction, validation, handoff to downstream systems.
Support & ops automation
Triage, routing, drafting. Fits existing tools. Biased toward escalation to keep humans in the loop.
How it typically goes
We start with the workflow, not the tech. What are people actually doing? Where does time go? What breaks? I need to understand the problem before proposing anything.
Then I build something real. Not a slide deck. Working software with instrumentation so we can measure what matters—accuracy, time saved, error rates, whatever the workflow demands.
Small scope first. A few documents, a single workflow, one team. Prove it works before expanding.
Then we iterate. Production systems need tuning. Edge cases surface. Requirements shift. I stick around to harden reliability and expand coverage as needed.
-
Agentic research system enriched 73,000 client records with industry-specific data. 98% cost reduction vs. manual lookup.
-
Document analysis pipeline extracts key dates and obligations from legal filings, flags contradictions, surfaces known operational pitfalls—with citations. ~10 hours saved per case at ~$15 compute.
-
FAQ automation across 200+ concurrent knowledge bases. Structured output to pre-approved responses—no manual configuration per case. Biased toward escalation to minimize risk.
-
Ticket triage routes high-priority items to PMs immediately, skipping the queue and reducing escalations.
Interested?
Tell me about the workflow, the current tools, the volume, and the constraints. I'll let you know if it's a fit.
George@ParksMC.com