Skip to content

Work

Systems we've designed and built, with the reasoning shown.

Lead intelligence, document intake, native AI products and connected operations. Explore the systems and the decisions that make them work.
  1. Lead workstation for synthetic record EX-1042, Cedarline Heating & Air, an HVAC contractor in Columbus, Ohio, marked Enriched. It shows the human-confirmed website cedarline.example with a candidate score of 91/100 across 3 ranked websites, two contacts (owner Morgan Ellis selected, service manager Taylor Reed as a candidate needing review), three site findings, an operator website-quality rating of 2/5 with CRM state "ready," and next-action buttons for Prepare outreach, Needs confirmation and Skip.

    Internal systemLead intelligence

    Scaling lead research without scaling manual research

    Varilead moves raw business records through identity resolution, website discovery, enrichment, qualification and outreach preparation, automating repetitive research while keeping identity and outreach decisions with people.

    173 operator hours a month released at 1,000 leads (modeled at $35/hour)

    View case study
  2. Diagram split by a vertical automation boundary. On the automated side: receive email and preserve source, parse and extract candidate values, validate and resolve with an operator, create structured CRM handoff. On the human side, fragmented pricing inputs and account context such as negotiated multipliers and legacy concessions feed the salesperson or estimator, who produces the commercially approved quote.

    Representative engagementIndustrial distribution

    Automating RFQ intake without automating commercial judgment

    Inconsistent RFQs had to be read, matched and re-keyed before pricing could begin. The proposed system combines deterministic parsing, bounded AI extraction and explicit review rules, while sales retains full authority over price, terms and exceptions.

    82% of RFQs routed to shortened review · ~38 hours a month released (constructed)

    View case study
  3. Three Velum screens: the dream archive, a saved dream with its summary, and the dream-work methods with done and in-progress states.

    Internal systemAI application

    Designing AI-assisted dream work as a product, not a chat window

    Velum turns saved dreams into guided, revisitable work through structured method sessions and persistent boards. Underneath the product, bounded model calls, explicit failure states and clear data boundaries keep AI inside the application rather than making it the application.

    Core Reflection + 6 guided methods · 11 board views · every model call authenticated, metered and logged

    View case study
  4. Two swimlane diagrams comparing the same lead before and after. Before: visitor, generic form, shared inbox, coordinator, a missing-information loop that calls the customer back, manual re-keying into Jobber, then the estimator, with a 2h 47m median staffed first response. After: visitor with UTM context, conditional form, property and service selection, deterministic ZIP check with an out-of-area decline, structured Jobber client and request, office notification and estimator, with a 26m median and a dashed failure branch that retries, then alerts and emails the office.

    Representative engagementHome services

    Turning a quote form into an operational lead pipeline

    A website redesign became a lead-operations project when discovery showed every inquiry stopped in a shared inbox. The rebuilt intake qualifies property class, service and location, then creates a structured Jobber request instead of leaving office staff to interpret, chase and re-key each lead.

    Median staffed first response 2h 47m → 26m (constructed)

    View case study
  5. Pinemark exception queue with synthetic orders: counts by state, filters, and seven exceptions with rule, order, age, owner, assignment SLA, source systems and status.

    Representative engagementEcommerce operations

    Giving ecommerce exceptions a durable operational lifecycle

    Shopify knew the order, ShipStation knew the shipment and Gorgias knew the customer conversation, but no system owned the exception. Future Built added normalized event state, deterministic rules and an operations console that turn each exception into owned work, without replacing any of the three.

    Time to owner 6h 12m → 1h 08m (constructed)

    View case study

Varilead and Velum are internal systems. Norwyn, Lakebridge and Pinemark are representative engagements, with constructed companies and results labeled in each case. They are not historical clients.