Selected Work

Business systems, with the decisions, data flow, and handoff explained.

Each write-up states the business problem, the system designed for it, and what that system verifiably does. Where a client result is not something we can evidence, it is not claimed.

Explore Selected Systems Discuss Your System

About these write-ups. They describe system behavior that can be demonstrated — what gets captured, where it is stored, how it is routed, and who it reaches. Client names, revenue figures, and percentage improvements are only shown where the client has approved them, which is why you will not see invented numbers here.

Overview

Selected systems

A closer look at how each one moves from trigger to verified operational handoff.

1 / 2

Automation system

Memory Preview — Human-Governed AI Knowledge System

A governed AI knowledge system that separates conversational reasoning from durable project memory. ChatGPT can propose structured knowledge updates through an MCP bridge, while provenance, project isolation, duplicate and conflict checks, and explicit human approval control what becomes permanent. Approved knowledge is versioned in GitHub and can be retrieved later as project-scoped context, preserving traceability and human oversight.

ChatGPTMCPGitHubAI AgentsHuman-in-the-LoopKnowledge ManagementAI GovernanceVersion Control

Discuss something like this

The business problem: AI conversations contain valuable project decisions, requirements, and lessons, but that context is often lost between sessions or stored without clear ownership, provenance, or review. The challenge was to create durable project memory without allowing an AI agent to write directly into trusted knowledge. The system also needed to prevent cross-project leakage, detect duplicates and conflicts, preserve source evidence, and require explicit human approval before any proposed knowledge became permanent.
The system designed: ChatGPT performs conversational reasoning and prepares structured knowledge proposals through a project-scoped MCP integration. The MCP server acts as the controlled bridge between ChatGPT and the knowledge repository, enforcing project authorization, schema validation, provenance requirements, duplicate and conflict checks, and approval workflows. Proposed changes remain drafts until a human confirms the exact approval challenge. Confirmed proposals open a draft GitHub pull request, where version history, review, and traceability are preserved. Compiled project knowledge can then be retrieved in later sessions through authorized, project-scoped context tools.
What it does in operation: Memory Preview gives AI-assisted projects continuity without sacrificing human control. Teams can preserve decisions, requirements, architecture choices, project state, and validated learnings across sessions while keeping every permanent update reviewable and traceable. Project isolation reduces the risk of knowledge leaking between clients or workspaces, provenance shows where information came from, and GitHub-based versioning provides an auditable change history. The result is a safer and more maintainable way to use long-term AI memory in real project workflows.

1 / 3

Automation system

Automated Real Estate Lead Qualification & Outbound Calling System

A three-workflow closed-loop system for real estate lead qualification: scheduled intake normalizes and deduplicates leads before dispatching authenticated outbound AI voice calls via Retell, inbound caller lookup matches phone numbers back to lead records, and post-call webhook processing analyzes transcripts, updates records, and routes qualified opportunities to human acquisitions handoff — with API response validation, failure routing, and monitoring throughout.

Retell AIClaude/OpenAIGoogle SheetsGoogle DocsGoogle DriveWebhookn8n

Discuss something like this

The business problem: Raw property leads needed consistent validation, qualification, calling, and follow-up without losing context between tools.
The system designed: Scheduled intake normalizes and deduplicates records, authenticated Retell calls handle outbound conversations, and post-call webhooks analyze transcripts before routing qualified opportunities to a human acquisitions queue.
What it does in operation: The build turns disconnected lead tasks into a monitored flow with explicit validation, failure paths, and a clear human handoff.

Automation system

Real Estate Wholesaling Buyer Intelligence

n8n deal-intake system that geocodes property addresses, checks a 15-day address cache, looks up property and buyer data, skip-traces LLC and person routes, normalizes contacts, and uses AI to rank the best cash buyers into Google Sheets.

n8nGoogle Geocoding APIBatchDataSkip SherpaOpenAI APIGoogle Sheets

Discuss something like this

The business problem: Buyer research required several location, property, skip-trace, and ranking steps that were difficult to repeat consistently by hand.
The system designed: An n8n workflow geocodes the property, checks a reusable address cache, gathers property and buyer data, normalizes LLC and person contacts, then uses AI-assisted ranking to prepare a focused buyer list.
What it does in operation: The system keeps data provenance visible while reducing repetitive research and presenting the team with a cleaner shortlist for review.

Automation system

Facebook Messenger AI Agent

Webhook-based Messenger agent that filters events, fetches context, generates AI replies, triggers downstream actions via HTTP, and responds back through the webhook flow.

n8nWebhooksAI AgentHTTP RequestGemini Chat ModelMemory

Discuss something like this

The business problem: Incoming Messenger conversations needed fast, context-aware replies without allowing an AI response step to control the entire workflow.
The system designed: Webhook events are filtered and enriched with conversation context, an AI agent drafts the response, controlled HTTP actions trigger downstream work, and the final message returns through the Messenger flow.
What it does in operation: Deterministic workflow steps retain control of routing and integrations while AI handles the language-heavy part of the interaction.

From proof to delivery

Have a workflow that should operate like a system?

We can map the trigger, rules, data, integrations, failure paths, and human handoff before deciding what should be automated.

Discuss Your System