LEADBASE AI

AI-powered B2B lead intelligence CRM for the Finnish market. Claude learns an ideal customer profile from your existing clients, translates it into Apollo.io search filters, and scores every prospect against an explicit rubric — each lead saved with fit signals, concerns, and a suggested outreach angle. Apollo-first search with Finland’s open YTJ registry as fallback.

Next.js 16SupabaseClaude APITailwind CSS

GitHub

Solo project · Full-stack & AI · 2026

03 / 05

Project Architecture

Technologies used

Next.js
React
TypeScript
Tailwind CSS
Supabase
Postgres
Deno Edge Functions
Claude API
OpenAI
Apollo.io

Team

Olli Karejoki

Teach it by example — your best clients in

Leadbase — client base settings page where the user teaches the system by example

One click — Claude generates the ICP

Leads ranked by fit — signals, concerns, outreach angle

Leadbase — saved leads ranked by ICP fit and score, each with signals and an outreach angle

The LLM is treated as an untrusted component

Scoring is not vibes-based: Claude applies an explicit point rubric derived from the user’s own ICP, and the server defensively parses, clamps, and re-validates every response before it reaches the database. A two-model cost split runs the constrained ICP-to-filter translation on Haiku and the judgment-heavy scoring on Sonnet. All external calls live in Supabase Edge Functions behind row-level security — the browser never touches a third-party API or key.

— Get in touch

Lets build something.

koodikommando@gmail.com→

GitHub