Selected Work.

Problem

Montgomery, Alabama’s city council allocates resources across nine districts serving over 200,000 residents using fragmented public data: crime reports, permits, 311 requests, business activity, and code violations, scattered across disconnected systems. Decisions end up reactive, shaped by political pressure rather than insight.

Approach

CivicPulse unifies that data into a District Health Index that scores each district across five weighted dimensions: public safety, economic vitality, city services, code compliance, and community access. Claude-generated council briefings translate the numbers into plain-language insights and recommendations.

Architecture

CivicPulse combines municipal data from Montgomery’s Open Data Portal with supplemental public information gathered through Bright Data. A data ingestion layer retrieves, normalizes, and enriches data before processing it through a custom scoring engine that generates a District Health Index for all nine council districts. The processed data is stored in Supabase and displayed through a Next.js dashboard, while Anthropic Claude generates plain-language council briefings based on district scores and supporting metrics.

The Result

Built and deployed a working CivicPulse prototype that evaluates all nine Montgomery council districts using live municipal data. The platform provides interactive district mapping, component-level scoring, district rankings, methodology transparency, and AI-generated council briefings that summarize district performance, identify emerging risks, and highlight areas requiring attention. Together, these features successfully transformed the fragmented municipal data into actionable intelligence for city leaders.

Problem

Smaller HVAC contractors often lose retrofit opportunities not because their pricing, equipment, or service is worse, but because they lack the sales and marketing resources to track incentive programs and turn them into compelling customer proposals. Large HVAC companies can dedicate teams to monitoring rebates, calculating energy savings, and building branded ROI-driven sales materials. Smaller contractors usually cannot. As a result, available incentives go unnoticed, customers are shown only the upfront installation cost, and stronger financial arguments are left out of the sales process.

Approach

RebateIQ was built as an AI-powered sales enablement system for independent HVAC contractors. The platform monitors rebate and incentive programs, identifies which opportunities apply to a contractor’s market, and helps turn those incentives into customer-facing sales material. Contractors can use RebateIQ to generate targeted outreach, schedule site visits, and create branded proposals that show customers the full financial case for upgrading their equipment. The goal was to give smaller contractors the same incentive intelligence and proposal capability as larger competitors, without hiring a dedicated sales or marketing team.

Architecture

RebateIQ is a five-agent system built on Google Cloud Agent Builder and Gemini, with a coordination layer passing work between agents that each have a distinct role. Elasticsearch, reached through Elastic’s MCP server, indexes known incentive programs and checks them against government and utility sources like DSIRE to flag new ones. Its semantic search matches an AI-written ideal customer profile against local property managers and other businesses to build a prospect list. Gemini then drafts a tailored outreach email in the contractor’s voice that highlights the program and its financial benefits. When a prospect replies, semantic search identifies interest and Gemini drafts a response. Available time slots for site visits are pulled from the contractor’s Google Calendar through the Calendar MCP. After the visit, semantic search matches the customer’s equipment to the programs it qualifies for, and the proposal agent fills a branded template with the results.

The Result

Built a working RebateIQ prototype with all five agents functioning: the monitor flagged new and changed programs, the prospect agent built a targeted list of prospects that would likely benefit from the rebate program, the outreach agent then drafted a tailored cold-email campaign, and the scheduler turned an interested reply into a booked visit. The proposal generator then turned site details into a company-branded PDF. The proposal demo used mock data for a 38-suite apartment retrofit, replacing a 72%-efficient boiler with a 95%-efficient system. From those inputs, it produced a proposal showing $2,712 in rebates from the Enbridge program, $3,719 in annual energy savings, and a 14.9-year payback.

Problem

A capable AI model can deliver the kind of insight a consulting firm charges thousands for, but only when it is given the right instructions and access to the right company data. Most business owners do not know how to set that up, and smaller companies cannot justify the cost of a consulting firm. So the value stays locked away.

Approach

This skill runs the engagement a consultant would. It researches the business, its industry, and its competitors, then takes the owner through a series of discovery questions to learn how the business makes money and where better data can sharpen decisions. From there it builds a tailored library of prompts, each naming the exact folder or software Claude needs to connect to, so every answer is grounded in real company data. When there is insufficient data, the prompts state what to start tracking for a more accurate analysis. The owner runs the prompts each month to turn their own data into consultant-level insights.