Full Stack Dev / AI-powered automation

Full-Stack Developer (Lovable + Supabase + AI Automation + Google Vision) — 3-4 Week Contract HVAC Lead Generation SaaS | Remote | Apply Immediately | Long-Term Potential The Project We're building the first AI-powered lead qualification platform for the $160B HVAC industry. Two applications already scaffolded inside Lovable — we need a full-stack developer to wire them to production and ship. Homeowners get screwed on HVAC quotes every day. Contractors oversell equipment using sizing "rules of thumb" that are 40+ years old. We built a system that stops it: the Guzzler Score — an 8-factor home efficiency rating that protects consumers and pre-qualifies leads before they ever hit a sales floor. Think Carfax for your AC. Here's how the funnel actually works: Phase 1: The Quiz (Preliminary Score) Homeowner answers 12 questions about their HVAC system — age, energy bills, repair history, hot/cold spots, noises, etc. The scoring engine (configurable weighted algorithm) outputs a Preliminary Guzzler Score from 0–100. Higher = worse. Immediately they see "78/100 — High Guzzler." Phase 2: Photo Verification (The Conversion Engine — THIS IS THE HARD PART) The preliminary score has a hook: "But your data plate might tell a different story than your memory." The homeowner is guided through a gamified 7-photo verification sequence across 4 sets by "Cora," our AI handholding agent: Set Photos Dollar Unlock What Google Cloud Vision OCR Extracts Set 1: Outdoor Unit 3 photos (stand-back, close-up, data plate) $325 SEER rating, tonnage, manufacture date, refrigerant type (R-22 vs R-410A), model number, brand, coil condition Set 2: Breaker Panel 2 photos (panel closed, breakers open) $100 Panel capacity, breaker sizing, dedicated HVAC circuit detection Set 3: Thermostat 1 photo $75 Thermostat type (smart/programmable/manual), staging capability Set 4: Electric Bill 1 photo $500 kWh usage, monthly cost, rate tier, utility provider, billing period Total unlock: $900 in discount credits. Progress bar moves across the screen. Cora congratulates between sets. Confetti on full completion. The verified Guzzler Score is revealed immediately — and it often differs from the preliminary score, which creates the trust moment: "They caught something I didn't know." This is NOT a simple photo upload form. It's a sequenced, stateful, gamified verification pipeline with: Google Cloud Vision OCR extracting structured data from HVAC data plates (model numbers, SEER ratings, manufacture dates — often tiny text on reflective metal stickers photographed by non-technical homeowners) Google Cloud Vision analyzing equipment condition from outdoor unit photos (rust, debris, coil condition) OCR on electric bills extracting kWh usage, rate tiers, and billing periods from wildly inconsistent utility formats Real-time score recalculation as photo data overrides self-reported quiz answers Skip handling: photos can be skipped with flags for in-home verification Safety guardrails: Cora warns homeowners not to touch electrical panels or enter unsafe areas Phase 3: Virtual Consultation & Close Verified-score homeowners book a 20-minute virtual call. The Guzzler Score PDF (4-page Energy Drain & Equipment Lifecycle Analysis) auto-generates from quiz + photo data and is auto-emailed on calendar booking. Our sales team closes at 90% on verified leads. Phase 4: Contractor Marketplace (B2B Monetization) The Guzzler Score powers a color-coded lead marketplace where HVAC contractors browse pre-qualified, scored leads: Tier Color Score Price to Contractor Service/Repair 🟡 Yellow 0–49 $45–75 Replacement 🟠 Orange 50–75 $150–250 (exclusive) Pre-Sold Appointment 🔴 Red 76–100 $400–600 (exclusive, 24hr claim window) Virtual Conditional Proposal 🟢 Green Any 10% revenue share Leads include the full diagnostic profile: Guzzler Score with subscore breakdown, Google Vision-verified equipment specs, electric bill analysis, EDS load calc sizing data, and Cora's conversational history. All leads are exclusive to one contractor per territory. What's Already Built Both apps scaffolded in Lovable (Pro plan, $20/mo) with full UI/UX Supabase connected at the project level (Postgres, Auth, Storage, Edge Functions available) Brand identity, design system, user flows, and quiz frontend are complete Guzzler Score algorithm is fully spec'd: 12 questions → 8 weighted subscores → preliminary score → photo-verified override → verified score Pricing model, funnel rules, marketplace logic, and photo verification sequence are documented (locked spec) Hermes AI agent framework deployed on the business side (Slack alerts, follow-up automation) What You'll Do (4-6 Weeks) Week 1-2: Database, Auth & Scoring Engine Wire Lovable apps to Supabase: proper schema, relationships, Row Level Security User authentication flows (email/password + Google OAuth) end-to-end Build the configurable Guzzler Score weighting engine (weights in a config file, not hardcoded — we will tune them as data comes in) Data models: quiz responses, preliminary scores, Guzzler Score assessments (8 subscores), user profiles, photo uploads, vision results, contractor listings, lead marketplace transactions, claim records, revenue share tracking EDS load calc API integration for the Sizing Mismatch subscore Week 2-4: Photo Verification Pipeline & AI Agent Integration (THE CORE) Google Cloud Vision OCR pipeline: Upload flow: homeowner submits photos → Supabase Storage → Google Vision API → structured data extraction → score recalculation Data plate OCR: extract SEER, tonnage, manufacture date, refrigerant type, model number from HVAC data plates (inconsistent formats, reflective surfaces, tiny text — this is the hardest OCR problem in the stack) Electric bill OCR: extract kWh usage, monthly cost, rate tier, utility provider from wildly varying utility bill layouts Equipment condition analysis: coil condition, rust detection, debris from outdoor unit photos Gamified verification sequence: 4-set progress bar with dollar unlocks, Cora transitions between sets, confetti on completion, skip handling Cora AI handholding agent that guides homeowners through both the quiz and the photo upload — conversational UX, prompts for photos, analyzes responses, explains scores in plain English Real-time score recalculation: when Google Vision returns verified equipment specs that differ from self-reported quiz answers, the verified Guzzler Score updates immediately Twilio SMS integration: verification reminders, follow-ups, contractor appointment confirmations Week 4-6: Marketplace, Payments & Production Contractor lead marketplace: browsing, filtering by score/tier/urgency, claiming with exclusivity rules (Yellow = shared max 2, Orange = exclusive first-claim, Red = exclusive + 24hr countdown timer) Dynamic pricing with Time-to-Talk (TTT) multiplier: ASAP leads get 1.5× surge pricing and instant SMS push to contractors Stripe payment integration for lead purchases (already provisioned in Stripe Projects) Slack webhook: real-time sold deal alerts and contractor claim notifications to the sales team Guzzler Score PDF auto-generation and email delivery on calendar booking Out-of-network flow: ZIP code lookup → optional $14.99 standalone PDF purchase for homeowners outside service area Production deployment, QA, edge cases Tech Stack Layer Tools Frontend Lovable (React/TypeScript output) Backend/Database Supabase (PostgreSQL, Auth, Storage, Edge Functions, Realtime) AI/Agent LLM API integration, conversational UX, prompt engineering Computer Vision / OCR Google Cloud Vision API — document_text_detection for data plates and electric bills, label_detection for equipment condition Communication Twilio (SMS), Slack API (webhooks) Payments Stripe (provisioned via Stripe Projects) DevOps Vercel/Netlify (Lovable deploy), GitHub You're a Fit If You: Have built and shipped full-stack apps with Lovable + Supabase — you know the quirks Strong React/TypeScript fundamentals (you can go under the hood when Lovable needs it) Experience with Google Cloud Vision OCR — you've extracted structured data from inconsistent document formats (data plates, utility bills, labels) Experience with AI/LLM integration — prompting, conversation design, agent UX, real-time data injection into agent context Have wired Twilio for SMS automation (transactional + conversational) Understand database design: schemas, RLS policies, Edge Functions, real-time subscriptions Can build gamified, stateful user flows — progress bars, staged unlocks, skip handling, confetti triggers Move fast — this is a 4-6 week sprint, not a 6-month engagement Strong preference for developers who have built OCR pipelines on user-submitted photos (receipts, documents, equipment labels) and/or gamified consumer onboarding flows. If you've shipped something in HVAC, home services, or insurance, you're hired yesterday. The Opportunity This is a 4-6 week contract to ship v1. But the platform has paying HVAC contractors waiting and a clear growth path. The right developer transitions to a long-term role: new features, scaling the marketplace, expanding to adjacent trades (plumbing, electrical, roofing), building the Guzzler Map (predictive permit-level HVAC health inventory for entire metro areas). You'd be the technical backbone of a funded, revenue-generating startup in an industry that's desperate for this. We move fast. If you're reading this and it fits, apply now. How to Apply 1. Start your proposal with the word "Guzzler" so we know you read the whole thing 2. Tell us about a Lovable + Supabase project you shipped 3. Tell us about an OCR pipeline you built — specifically, extracting structured data from inconsistent, user-submitted photos (receipts, labels, documents, equipment plates) 4. Share one thing you'd improve about the gamified photo verification experience for a non-technical homeowner 5. Your availability: can you start this week? Budget: Fixed price — $1,000 to $1,500 for the full 4-6 week scope. Flat fee, milestone-based. No hourly. Here's the breakdown: Milestone Deliverable Payment Milestone 1 Supabase schema, auth flows, RLS policies, configurable scoring engine wired to both Lovable apps $350-500 Milestone 2 Google Vision OCR pipeline (data plate + electric bill extraction), Cora AI agent with conversational UX, gamified photo verification sequence (4-set progress bar, dollar unlocks, confetti, skip handling), real-time score recalculation, Twilio SMS flows $350-500 Milestone 3 Contractor marketplace live (color-coded tiers, exclusivity rules, 24hr countdown timers), TTT dynamic pricing, Stripe payments, Slack alerts, PDF auto-generation, production deploy $300-500 Tell us your fixed price for the full scope. If you need to see the apps first to quote accurately, say so — we'll screen-share inside Lovable. No hourly. No scope creep. Milestone releases on delivery. ComfortIQ — The Guzzler Score. 12 questions. 7 photos. One verified diagnosis. Protect the homeowner. Pre-qualify the lead. Close at 90%.