Skip to content

Product studio · Paris

What costs you two days a month can run itself in six weeks.

We build internal tools that ship to production, not mockups. We know how because we already run our own, with real customers and real outages.

A written answer within 24 to 48 hours, including when it is no.

What we deliver

A deployed app, a link you can use, one main workflow running.

How long it takes

A few weeks. Written scoping before we start.

What we do not do

Time and materials development against a vague spec.

Who for

Two situations we know inside out

Digital SMB · 15 to 60 people

Agencies, SaaS vendors, established e-commerce

You have data in three tools that ignore each other, and someone who spends every Monday bridging them by hand.

  • Re-entering data between CRM, invoicing and delivery
  • Client reporting rebuilt by hand every month
  • A key process that lives only in a spreadsheet

Training provider · Qualiopi

Professional training, 10 to 50 people

Certification demands evidence, evidence demands paperwork, and paperwork eats the margin.

  • Building and tracking Qualiopi evidence
  • BPF and regulatory reporting rebuilt by hand
  • Assessments, placement and learner tracking scattered

Work

What we build, and what we keep standing

Four systems, four different constraints. Each page tells how it was built, not what it does: that is its own site's job.

Notacor

In production

An intelligence layer on top of a notary firm's data.

Off-schedule opportunities among dormant clients, argued one by one. A database isolated per firm, professional-conduct guardrails in the code, sovereign hosting.

Making of →

Blindspot

Sold online

Is your brand cited when an AI answers in your place?

Unprompted visibility measured on implicit prompts, across four models. A self-serve product that takes payment and delivers.

Making of →

Andiamo

In production

A travel comparison tool that lives inside ChatGPT and Claude.

Train, coach, plane and car compared door to door, from a single sentence. Nothing to install: the product plugs into the assistant already in use, so it has no screen of its own.

Making of →

Leon

Engineering note

An internal test bench: keeping a language model inside an irreversible decision chain. Neither a product nor a published benchmark. Worth reading if you are considering putting AI on a process that genuinely commits something.

Read the note →

What it proves

These products are not a portfolio. They are our production track record.

Building for yourself forces what a vendor can avoid: running it, fixing it, paying for your own mistakes. Here is what each constraint made us learn, and what it gives you.

OriginThe constraint, and what it demandsWhat already runs here

Notacor

The constraint

Sensitive professional data. Demands isolation that holds even through a bug, and a scope bounded from the data model up.

What runs

One PostgreSQL schema per firm, bounded enrichment workers, hosting and AI inside the EU.

Notacor

The constraint

Non-negotiable professional conduct. Demands business rules written in code, readable and testable one by one.

What runs

Targeting engine and guardrails in a pure domain package, independent of the interface.

Blindspot

The constraint

A product sold online. Demands payment, quotas, API costs held in check, and a defensible measurement.

What runs

Three plans on sale, a score based on implicit prompts, four models queried.

Andiamo

The constraint

No screen: the interface is a language model. Demands that data carry its own degree of certainty, and that whatever costs money be bounded per user.

What runs

Route composition server side, one score shared across modes, a "published" or "estimated" status on every leg, quotas per account.

Leon

The constraint

Automatic decisions with real money. Demands wrapping the probabilistic in the deterministic, and being able to stop everything.

What runs

Deterministic validation with the last word, structured features, idempotency, a dead man's switch.

When we deliver for you, you do not inherit a method. You inherit systems we already keep standing.

One case

4 hrs became 20 min

Per case file, measured before and after going into production

Building a case file went through three tools and two rounds of re-entry. The flow was rebuilt as a single screen, resumable from one session to the next, with consistency checks run at entry time rather than at review.

Head of operations, consulting firm, 30 people

Foundations

Two building blocks we can deploy or adapt

See all →

Finova

Structured client onboarding for wealth management firms: the file comes out usable, not ready to retype. Roughly 2 to 4 hours saved per file.

finova.looptra.ai ↗

Digital Twin

A chatbot trained on a resume, a bio, a few links. It introduces you in your place, through a shareable link.

digitaltwin.looptra.ai ↗

Method

Four steps, nothing more

Step 01

Identify the friction point

A measurable symptom, not an idea.

Step 02

Qualify

Within 24 to 48 hours: addressable or not, and on what terms.

Step 03

Ship to production

A few weeks, one main workflow, a link you can use.

Step 04

Iterate

Short cycles, with the people who use it.

Working with us

A first version, deployed to production. Not a mockup.

Scope and terms are set after scoping, in writing. We would rather give you an accurate number once we have listened than a price grid that means nothing.

What is included

  • An answer on feasibility within 24 to 48 hours
  • Written scoping: scope, main workflow, roadmap
  • A deployed app, reachable by a link, used by real people
  • Iterations included after the first field feedback

What is not

  • Several workflows and personas in parallel
  • Advanced roles and permissions
  • Invoicing, subscriptions, payment
  • Ten integrations and real-time sync

Need that? It is an iteration, not a first version. We plan it in.

Contact

Tell us what costs you time. We answer within 24 to 48 hours, including when the answer is no.

A concrete symptom beats a spec document: what gets redone by hand, what breaks, what gets lost between two tools. We tell you whether it is addressable, how we would break it down, and what we would do first.

  • A written answer, from the person who will build it
  • No automated sales sequence behind it
  • If it is not for us, we say so and explain why

contact@looptra.ai

12 rue Letellier, 75015 Paris

An answer within 24 to 48 hours · No commitment