Independent AI systems builder · Omaha, Nebraska
I turn documents, workflows, and operational knowledge into inspectable AI systems — with guardrails, provenance, and a human in control.
Working systems,
not slideware.
Selected work
Each project solves a real systems problem: retrieval, orchestration, safety, memory, or the operating layer around them.
A RAG chatbot with a genuine tool-calling agent on top — it searches uploaded documents, reports knowledge-base stats, and generates charts, all decided by the model itself. Swappable LLM provider (OpenAI, Anthropic, or OpenRouter).
knowledge_base.searchcomplete02chart.generatevalidatedProject evidence
Real interfaces and command output from local runs — no concept screens or invented dashboards.
Open full capture ↗The real operator-facing overview: hybrid retrieval, scoped memory, lifecycle governance, and explainability.
Repository
Open full capture ↗A current 111-project scan with candidate scoring, skills coverage, release gaps, and human-reviewed next steps.
Repository
Open full capture ↗The real FastAPI contract with provenance-aware statement ingestion expanded in the OpenAPI interface.
Repository
Open full capture ↗Actual CLI output for project documentation, diagrams, reports, GitHub setup, and work-session tracking.
RepositoryA graph-native memory layer for AI agents — hybrid vector, fulltext, and graph retrieval, with a full memory lifecycle (ingest, reflect, promote, archive) instead of stuffing everything into a prompt.
AI-assisted network monitoring and remediation for OpenWrt — log analysis, settings audits, threat detection, and safe command-execution gates before anything touches production.
Scans my own project portfolio for reusable proof, scores product and market opportunities against what I’ve actually built, and turns job leads and market signals into a prioritized action queue.
A modular, project-agnostic runtime for AI agents — model routing, tool execution, policy enforcement, and child-agent delegation, so an application only has to own its domain logic, not execution mechanics.
An event-sourced, bi-temporal knowledge graph kernel for AI agent memory — immutable assertions with full retraction lineage, epistemic reasoning, hybrid retrieval, and multi-hop pathfinding.
A local-first CRM for freelance pipelines — a Kanban board over SQLite with suggest-and-confirm automations: stale-lead alerts, LLM-drafted follow-ups, and stage-change suggestions. Nothing moves without a human confirming it.
A developer workflow CLI — work-session tracking, auto-generated docs and architecture diagrams, and GitHub repo scaffolding, backed by SQLite instead of loose files.
Ways to work together
No open-ended transformation pitch. Choose the smallest engagement that can produce useful evidence and a responsible next decision.
Find the part of a workflow where automation would actually remove friction — and where it would only add risk.
Turn a real document set into a private, citation-backed assistant that shows how every answer was formed.
Connect retrieval, tools, APIs, and approval gates to an existing product or internal workflow.
Working rule Every engagement keeps sources visible, decisions inspectable, and consequential actions behind explicit approval.
How I build
Ground the system in the documents, workflows, and facts the business already trusts.
Make retrieval, routing, tool use, and failure states visible enough to inspect and improve.
Put validation and approval boundaries between a confident model and a consequential action.