AI Feature Sprint

One AI feature, properly built,
in 2–4 weeks. Fixed price.

Search that finds the right answer. A chatbot that doesn't make stuff up. A summary that's actually useful.

What we build

Smart Search

Search that understands what users mean — not just what they typed. Finds the right answer even when wording differs.

pgvector · Cohere Reranker · BM25

Document QA / Chat-with-PDF

A chatbot that answers from your own documents — and shows where each answer came from. No making things up.

LangGraph · Vercel AI SDK · pgvector

AI Insights / Summary

Automatic summaries and insights for your dashboards. Updated on a schedule, structured for easy display.

OpenAI · Anthropic · Promptfoo

Smart Onboarding

An AI assistant that guides new users through setup — adapting to what they actually need.

Vercel AI SDK · OpenAI

Variants
FeatureLight €5,500Full €12,000Multi-Agent (Custom)
Number of features1 feature1 featureComplex multi-agent
Answer appears live (word by word)
Shows where answers come from
High-quality searchoptional✓ default
Quality measurement system✓ workflow-level
Test set with real questions
Automated quality checks
Protection against failuresbasicfullfull
Cost tracking
Insight into how the AI behaves
Bug-fix warranty14 days30 days30 days
Handover meeting90 min90 min
Timeline2 weeks3–4 weeks4–6 weeks
Price€5,500€12,000€22,000–35,000

The Multi-Agent option is a custom build — start with an Audit or Strategy Sprint first so we know what to scope.

For your CTO — what we'll use
For your CTO — what we'll use

Default stack: Next.js, TypeScript, OpenAI or Anthropic for the AI brain, Postgres with pgvector for storing embeddings, optional Cohere or BGE rerankers for higher-quality search, Vercel for hosting. Quality measurement: RAGAS or Promptfoo, with LangSmith or OpenTelemetry for tracing. For Multi-Agent setups we pick the framework that fits your stack: Mastra or the Vercel AI SDK for TypeScript shops, Pydantic AI or LangGraph for Python, AWS Strands for AWS-native teams — with a durability layer (Temporal or Vercel Workflows) when a workflow runs for minutes or longer. Chosen per use case, never one-size-fits-all.

Why this won't break in 6 months

Most AI projects ship a demo that works on Tuesday and starts misbehaving on Friday. Models change, costs change, your data changes — and nobody notices until users complain.

We do it differently.

Before we write any code, we agree on a list of real questions your users would ask and the answers they should get. We test against that list while building. After launch, the same tests run automatically — every time the AI changes, every time the model updates.

If quality drops, you'll know on day one — not after your first angry customer email.

Process
Week 1

Plan it. Agree on what "good" looks like with a real test set.

Week 2

Build the core feature.

Week 3

Add quality checks, cost tracking, protection against failures.

Week 4

Polish, hand it over, document everything.

What's NOT included
  • Per-customer quality testing if you have a multi-tenant product (separate add-on)
  • Custom model fine-tuning
  • On-prem deployment (separate Self-Hosted Add-On, +€3,500)
  • Rebuilds of existing non-AI features (see MVP Sprint)
Self-Hosted Add-On

Need to host the AI on your own servers — for GDPR, compliance, or because your enterprise customers demand it? We add that on for €3,500. 1–2 weeks extra.