# Preste

> Preste is a reviewed AI agency in Paris, France, listed in the AIAgencies.eu register and last assessed 2026-09-24.

Canonical page: https://www.aiagencies.eu/agency/preste
Generated: 2026-09-28 · Curated by Gabor Kiss · No paid placements — inclusion, score, and rank are never sold.

## Facts
- Location: Paris, France
- Website: https://www.preste.ai
- Hourly rate: €45-85/hr (directory-listed)
- Team size: 10-49
- Founded: 2019
- Services: AI Development, Generative AI, AI Consulting
- Industries: Healthcare, Manufacturing, SaaS & B2B
- LinkedIn: https://www.linkedin.com/company/preste

## About
Preste is a Paris AI firm founded in 2019, working across computer vision, anomaly detection, document processing, conversational agents, and molecular design. It is an NVIDIA embedded software partner building on Jetson platforms.

## Documented clients
Documented engagements: Greenbig, Streamowl, Tecbak, Idemia, Invofficine, Espaces, UNAPEI, Smartwatt, Cinétévé, INSERM, Sciences-Po Paris, and Siemens Healthineers. All 12 carry a source URL on file.

## Curator assessment
Reviewed 2026-09-24 by Gabor Kiss against published evidence.

Preste's achievements page carries 17 project cards that name the client, the objective, the stack and the build time: a Jetson Orin Nano pipeline embedded in Greenbig's B:Bot recycling machine, a YOLOv8 thermal-leak detector mounted on Streamowl's drones, a TensorRT model reading foosball games at 120 FPS for Tecbak, and document extraction inside Invofficine's pharmacy software.

No card reports a result or says whether the system is still running, and three describe one-month design, specification or test work (Mayam, MadMonkeys, Compiègne University), so ask which builds are in production today.

The weaker fit is buyers who need a large LLM product team: the language work is real—RAG search for Espaces, LLM-assisted document simplification for UNAPEI—but the depth, and the NVIDIA partnership, sit in embedded vision.

### Key strengths
- Stack and hardware published per project—Jetson Orin Nano, Xavier and TX2, TensorRT, DeepStream, Hailo-8—with build times from 1 to 10 months
- Research record checkable off the site: Preste's part in two Horizon Europe consortia (NOVA 2022–2026, TOXBOX 2024–2028) and a joint paper with INSERM
- Engineering blog publishes method and numbers, including an unsupervised visual-inspection test (mean IoU 0.543) and a synthetic-data NLP experiment with its code on GitHub

### Good to know
- No card states an outcome or production status—ask for two references whose systems are running today and what they measure
- There is no team page; the CTO and a few blog authors are the only named people—ask who staffs the build and how the Paris, Prague and Kyiv offices split the work
- Preste publishes no rate itself; the €45–85/h shown here comes from a third-party directory listing—confirm whether edge hardware and post-launch maintenance are quoted separately

---
Source: AIAgencies.eu — the curated register of European AI agencies. Rate provenance: "agency-confirmed" = disclosed via submission or claimed profile; "directory-listed" = Clutch or two agreeing B2B directories (lone unverified sources are never written). Full method: https://www.aiagencies.eu/methodology · Rate index: https://www.aiagencies.eu/rates