Skip to content
— KCG Products · Pillar four —

We don't only consult. We ship.

Production-grade AI products we built for ourselves and our markets. Because we don't trust an AI consultant who's never shipped one of these. Each is a live demo — click and break them.

Build yours with us or — try the live demos
— Five production products —

Click. Break. Tell us what you broke.

01 — SKARB.AI
Skarb — family archive RAG
A trilingual family-archive app where users converse with their own memories. Production GraphRAG architecture, web + iOS, Stripe billing, real paying users. The deepest technical product we have shipped.
voyage-3-large · pgvector + HNSW · Supabase · Capacitor
Visit Skarb.ai
02 — SCHOOL
Marketing OS for premium schools
A 12-module marketing operations platform for premium international schools. LLM-first — Claude Sonnet with tool use over Supabase. Twelve canned demo queries you can poke right now.
Claude Sonnet · tool use · Supabase · SSE streaming · React
Try the live demo
03 — ESTIMATES
Operations OS for renovation SMEs
Operations automation for non-technical construction SMEs. Bank-statement categorisation, estimate auto-generation, procurement agents, client-chat triage — solves the pains that nobody talks about because they are "just admin".
GPT-class LLM · Python · n8n · Postgres · automation
Try the live demo
04 — TENDERS
Tender + financial planning for tradespeople
A bidding-workflow and income-modelling tool for sole-trader electricians and contractors, built around real Polish-market tender standards (kosztorys, Norma STANDARD 2, ATH2).
GPT-class LLM · XML roundtrip · structured outputs
Try the live demo
05 — RETAIL
Analytics OS for multi-store retail
A retail-analytics workbench for multi-store operators: sales, margin and revenue-per-m² across locations, auto-generated insight narratives, scenario planning, and one-click CSV/Excel import.
Supabase · rule-based insight engine · Recharts · CSV/XLSX import
Try the live demo
— Why we build our own —

Eating our own cooking.

Most AI consultants have shipped nothing to production with their own name on it. They've shipped demos, slides, PoCs. When you hire them, you're paying them to learn what shipping production AI actually feels like.

We won't be those consultants. Every KCG product is a forcing function — it teaches us what we'd otherwise have to learn on your dime. An eval harness that's killed at least one approach per quarter. A retrieval pipeline that survived contact with real users. A pricing surface that converts. All five products are public, all five are real, all five are still maintained.

If you want help building one, the five above are also our reference architectures.