Classification
Domain-specific classifiers
Trained on your taxonomy: products, tickets, contracts, claims, leads. Higher precision than a generic LLM, lower cost per call.
Tuned to your data
Service · Custom AI
Built for your data, not the average user.
When the off-the-shelf assistant doesn't know your domain, your taxonomy or your customers, we build the model that does. Classifiers, extractors, scoring and retrieval — tuned to your data and run where you need it.
What it is
Generic AI ends.
Yours starts.
Off-the-shelf tools cover the easy 70%. The remaining 30% — where your real differentiation lives — needs models trained on your data, evaluated on your cases and run on infrastructure you trust.
We build that 30%. Sometimes it’s a fine-tuned classifier; sometimes it’s a retrieval system that respects permissions; sometimes it’s a small open-weights model running on your own servers. We pick what fits — not what looks impressive in a deck.
What we build
Six shapes of custom AI.
Extraction
Specialized document extractors
Forms, contracts, technical specs, multi-page reports. Structured output ready for the database, with confidence scores per field.
Field-level confidence
Scoring
Predictive scoring models
Lead quality, churn risk, claim severity, demand forecasting — calibrated on your history, with the reasoning exposed.
Explainable by default
Retrieval
Private RAG systems
Your documents, your permissions, your hosting. Retrieval that respects access rights and freshness, not a generic vector dump.
Permission-aware
Models
Fine-tuning & adaptation
When prompting hits its ceiling: fine-tune an open-weights model on your tone, jargon and edge cases. Run it on private infrastructure.
Open-weights friendly
Evaluation
Custom evaluation harness
Test sets built from your real cases. Regression checks on every change. Quality you can show the auditor, not just the user.
Continuously measured
How we ship it
Honest ML, no shortcuts.
Most “AI projects” fail at the dataset, not the model. We spend the time on the data, the labels and the eval — because that’s what holds up after launch.
01
Frame the problem
We pin down the inputs, the outputs, the accuracy threshold that matters and the cost ceiling that has to hold.
02
Build the dataset
From your historical data, with the team that knows the domain. Labeling protocol, gold set, blind eval set — the unglamorous work that makes models work.
03
Pick the smallest thing that works
Prompting, classical ML, embedding similarity, fine-tuning — chosen for the problem, not the trend. The simplest option that hits target wins.
04
Operate, monitor, retrain
Drift monitoring, quality dashboards, retraining cadence. A model in production is not a model finished — it's a model that needs a home.
Other services
AI automation
Repetitive back-office workflows running unattended, with audit logs and human review.
Read more →AI agents
Conversational, multi-step assistants for support, sales and internal knowledge.
Read more →CRM/ERP integration
Connecting AI to Salesforce, HubSpot, SAP, Holded, Odoo and the rest of your stack.
Read more →AI code audit
Independent senior-engineer audit of code built with Cursor, v0, Lovable, Copilot or any AI tool.
Read more →Trust
Safe, traceable AI,
enterprise-ready.
We design for privacy from the start, human control, traceability, usage limits, permissioning and documentation. For sensitive processes, we help assess risk and applicable obligations under GDPR and the EU AI Act.
- 01We never train models on your data without explicit authorization.
- 02Human review built-in for processes where risk demands it.
- 03Traceability: prompts, sources, permissions, errors and metrics — documented.
- 04Privacy, security and control integrated from day one.
- 05Solutions engineered to be maintained, audited and improved over time.
GDPREU AI ActAEPDISO 27001 readyEU data residency
Personal diagnosis
We work with
few clients.
Every engagement is led personally by one of the partners. If there's a fit, you get a personal first read of your case within one business day — not a canned demo.
How we work
- 01Tell us which process eats your time
- 02Personal reply within one business day
- 0320-minute call — no demo, no pitch