Analyze every valid inquiry immediately.
Building a resilient AI lead operations workflow.
LeadFlow AI is a production-minded portfolio project that explores how service businesses can respond faster without hiding important decisions inside a black box.
Product designAI automationTechnical operationsFull-stack development
01 / PROBLEM
Good leads are lost in operational gaps.
Service businesses often capture inquiries across forms, ads, referrals and email. The data is inconsistent, qualification is subjective, and follow-up depends on someone noticing the right message at the right moment.
The central problem is not merely lead volume. It is turning noisy input into a consistent, explainable next action—quickly enough to matter.
Design one continuous path from inquiry to action.
Show why a lead received its score.
Keep operating when an AI provider is unavailable.
Turn analysis into a specific human next step.
A CRM-shaped operations system with an AI boundary.
The product combines validated lead intake, deterministic or provider-backed analysis, a visual pipeline, configurable rules, and transparent activity history.
Capture layer
React Hook Form + Zod normalize structured lead data.
Analysis layer
An OpenAI-compatible adapter returns validated JSON.
Operations layer
Rules route, respond, notify and schedule follow-up.
Experience layer
Pipeline state, analytics and an auditable activity log.
The full lead lifecycle is visible.
01Capture→
02Validate→
03Analyze→
04Score→
05Route→
06Respond→
07Follow up→
08Learn
Every transition produces structured state that a human can inspect. AI recommends; business rules decide what happens next.
Built with the technologies the product actually uses.
Application
React 19, TypeScript, Next-compatible App Router
Interface
Tailwind CSS, shadcn-style components, Lucide
Data & forms
React context, browser demo persistence, Zod, React Hook Form
Automation
OpenAI-compatible API boundary with deterministic fallback
Analytics
Recharts and derived pipeline metrics
Quality
ESLint, Vitest, strict TypeScript, responsive accessibility
06 / CHALLENGES
Product decisions that required real tradeoffs.
Useful AI without fragile dependence
A deterministic scoring engine keeps the demo fully functional, while the same response contract supports an external model.
Realistic state without fake business claims
Seeded data demonstrates the workflow but all metrics are clearly illustrative—not presented as customer results.
Automation without a black box
Reasoning factors, activity events and human-readable rules make operational behavior inspectable.
Where the system could go next.
Supabase authentication, row-level security and multi-tenant workspaces
Durable database persistence and background job orchestration
OAuth-backed Gmail, Slack and HubSpot provider adapters
Webhook retries, idempotency keys and dead-letter handling
Prompt evaluation, model observability and human feedback loops