Products

What we build for ourselves

Consulting hours prove you can be hired. Shipping products that people pay for proves you can finish. We do both, deliberately — the engineering problems we solve in our own products are the ones we solve faster in yours.

KithFlow

Live

UK company intelligence, built for agents to consume

kithflow.com

A B2B intelligence layer over 5M+ Companies House records, exposed over both MCP and REST so an AI agent can query it as a first-class data source rather than scraping a UI.

The engineering

Every claim in a generated dossier is provenance-tracked back to its source record — a direct answer to frontier models confidently inventing company facts. A scheduled intent pipeline runs every six hours against user targeting rules.

  • Next.js
  • MCP
  • PostgreSQL
  • Anthropic
  • AI Gateway

HyperSaaS

Live

Django + Next.js SaaS foundation

hypersaas.dev

A production SaaS boilerplate covering the parts every product needs and nobody wants to build twice: auth, billing, organisations, background jobs, and document ingestion.

The engineering

Hybrid retrieval fusing pgvector cosine search over HNSW with PostgreSQL full-text via Reciprocal Rank Fusion, fed by a Celery ingestion pipeline with structure-aware chunking and citation-ready output.

  • Django
  • Next.js
  • Celery
  • pgvector
  • S3

KognitivAtlas

Live

An AI travel planner that actually writes the plan

kognitivatlas.com

Trip planning where the agent does not just suggest — it writes itineraries, day plans, and saved places directly into your trip, then keeps them current.

The engineering

Agentic write tools over live data sources: flight and hotel search, geocoding, weather and forecast, with usage-metered Stripe tiers so inference cost tracks revenue.

  • Next.js
  • Vercel AI SDK
  • Anthropic
  • Stripe
  • PostgreSQL

LLM-Wiki

Open source

A knowledge base an agent operates

llm-wiki.ozanyildirim.me

An MIT-licensed Obsidian vault template where Claude Code is the operator, with defined workflows for ingesting, querying, linting, and promoting notes.

The engineering

The instruction file is the product. Six deterministic agent workflows over plain Markdown — no server, no database, no lock-in.

  • Markdown
  • Claude Code
  • Obsidian

Why this matters to you

We pay for our own mistakes first

Running products on our own budget forces decisions that consulting work lets you avoid. Cost metering, provider fallback, retrieval quality and eval discipline are not theory here — they are things we got wrong on our own bill before we got them right on yours.

Next step

Tell us what you’re trying to build

Most engagements start with a fixed-price audit, so the first thing you buy is a decision rather than a commitment.