Franco Celeste

Problem

Specialist firm with strong credentials and no repeatable pipeline: no outbound engine, no CRM intake, no way to capture website visitors.

Build

Compliant cold-email infrastructure on warmed dedicated domains (SPF/DKIM/DMARC, CAN-SPAM unsubscribe verified on every template), HubSpot CRM intake, and a Claude-powered website chatbot on Cloudflare with hard monthly spend caps and fallback behavior. Daily launch surveillance with a written health brief.

Outcome

Launched July 2026 and operating live under a managed send ramp. Results earned a follow-on program proposal spanning a 12-bot expansion roadmap.

Problem

Six-person exotic rental operation across two cities, spending heavily on a sales process nobody had ever measured. 71% of it turned out to be manual.

Build

Full workflow-costing diagnostic: labor, revenue leakage, ad waste, and rework modeled step by step; six fix opportunities mapped, four of them no-code and startable within a week. Delivered as a board-ready report in 48 hours.

Outcome

$134,000/yr total workflow cost baselined; $74,230/yr net savings mapped (55% projected reduction); $61,000/yr in lead leakage surfaced.

Problem

Brokers spend hours drafting offering memoranda and deal analyses by hand; existing tools don't produce client-ready drafts.

Build

Sole architect and builder of a subscription web app: Cloudflare Workers + D1, magic-link authentication, Claude-powered document generation with pinned model versions, PDF pipeline with email delivery, Stripe subscriptions with customer portal, tiered usage caps with burst limits.

Outcome

Live product with self-serve checkout. Passed a formal pre-launch legal-claims audit (FTC click-to-cancel, CAN-SPAM, state consumer-protection statutes) — every customer-facing claim rewritten to survive scrutiny.

Problem

Boutique studio running onboarding, retention, and win-back communication entirely by hand, with no consistent follow-up.

Build

Complete customer-journey automation on the studio's booking platform: trigger design, audience filters, delays, channel selection, exit conditions, and message copy written to match the owner's documented brand voice.

Outcome

Delivered, approved, and paid in full, with a 30-day support window and a results-report close-out. Copy approved by the owner without voice corrections.

Problem

Business websites either have no conversational layer or an unbounded one that hallucinates claims and runs up API bills.

Build

Claude-backed chatbot on Cloudflare (Worker + D1) with versioned system prompts restricted to documented case figures only, later upgraded to a real-time conversational video avatar with minute caps and automatic text-chat fallback when limits are reached.

Outcome

Live in production since May 2026. Zero fabricated-claim incidents by design: the prompt cannot cite numbers that don't exist in source documents.

Problem

Industry surveys report that while 76% of developers use or plan to use AI tools, only 29% trust AI output accuracy (Stack Overflow). The norm is autocomplete and chat; unverified AI output is the bottleneck.

Build

A governed multi-agent operating system on Claude Code that runs the entire practice: 8 specialized agents under a documented org chart with explicit handoff contracts (strategy, sales, adversarial review, debugging, live client surveillance, session lifecycle) and 25 custom skills. A two-gate ship pipeline verifies every client-facing figure against source-of-truth files, then runs a separate 8-point brand/voice/scope check. Persistent cross-session memory with scheduled pruning audits, automatic pre-edit backups, secret/PII leak scans before every deploy, a monthly 4-sweep self-audit, and encoded governance: reversible decisions delegated to AI within defined impact thresholds, irreversible or external actions gated to the operator.

Outcome

Trust is not required, because outputs are verifiable by construction. One operator delivers diagnosis, build, launch, sales, and daily client surveillance at agency scale: every engagement above (CS-01 through CS-05) was produced through this system.

Maturity ladder is a self-assessment framed against published adoption data (Stack Overflow developer surveys); it is an architecture comparison, not an external certification.