2026/AI/In Development

Dealflow AI

AI-native CRM for freelancers — deals, tasks, and follow-ups in one place.

Personal product: Next.js web + Expo mobile sharing a Supabase backend. Pipeline for clients and deals, activity timeline, and AI that summarizes deals, extracts next actions, and drafts follow-ups — with an offline demo mode so the full UI works without env keys.

Web + mobile, one backend
AI summary → approve → tasks
Offline preview with demo data

The Problem

Freelancers juggle leads, proposals, and deals across spreadsheets, email threads, and half-adopted CRMs — none of which are built around how a solo operator actually works, or use AI to cut the manual tracking overhead.

What I Built

Dealflow AI is a personal AI-native CRM: a Next.js web app and Expo mobile app sharing one Supabase backend.

Clients, deals (kanban + list), tasks, activity timeline, and dashboard KPIs.
Deal workspace with notes, reminders, and file uploads (Supabase Storage).
AI that summarizes a deal, extracts next actions, drafts follow-ups, and can transcribe audio — with an approve → create tasks path so suggestions don’t silently mutate the pipeline.
Server-only /api/ai/* routes; mock adapters when keys are missing.
Offline preview mode — full UI + demo data with no .env, so the product is reviewable without infrastructure.

Major Features

Dual clients, one data model — web and mobile stay in sync through Supabase + shared packages.
Pipeline that matches freelance work — clients → deals → tasks → activity, not enterprise CRM bloat.
AI in the workflow — summarize, extract actions, draft follow-ups; human approval before tasks land.
Offline demo mode — ship and review the UX without wiring secrets first.
Typed monorepo — @dealflow/shared + @dealflow/ui keep web and mobile consistent.

Key Challenges

Sharing a clean data model between Next.js and Expo without a custom API layer, while keeping AI calls fast enough to feel like part of the deal workspace — not a separate chatbot tab. Mockable AI adapters were required so demos never depend on live keys.

Results

Dealflow AI is in development as a mobile + web AI CRM: core pipeline screens, AI action flow, and offline preview are in place. Roadmap items (email/calendar sync, push, offline write queue) are still open.

Stack Decisions

Supabase was chosen so both clients get typed access to the same Postgres data without maintaining a separate API. Server-only AI routes keep keys off the client; OpenRouter / OpenAI / Groq give flexibility for text and transcription without rewriting the product surface.

Screens