Custom AI agents, cost-effective and reliable in production

The manual processing of tens of thousands of CEE files was a major impediment to business growth. We developed Athos, an AI-First platform based on an agent system combining OCR and LLM: it classifies files, verifies their regulatory compliance, and isolates anomalies for instructors. The result: a bottleneck transformed into an industrialised flow, refocusing human effort on added value.

Mercialys faced numerous time-consuming operational processes and wished to define its AI strategy. We framed the end-to-end approach: establishment of a steering committee, facilitation of a roadmap of projects prioritised by ROI, then deployment of the first four agentic solutions in six months - contract analysis, internal chatbot, investment memo generation, and automated monitoring. A concrete, measurable transformation rooted in the teams' actual practices.

Manually reading and summarising complex tenders required numerous hours of work per consultation. We developed a bespoke AI agent capable of extracting technical and regulatory requirements from documents spanning several hundred pages, and automatically detecting inconsistencies. The analysis time is now drastically reduced.

Finding a specific clause or contract among over a million documents was a daily challenge for Gide's 500 lawyers. We developed and deployed a generative AI platform allowing searches across the entire corpus, and automatically extracting key metadata from each document. What previously required tedious manual exploration now becomes a query of a few seconds, with precise and contextualised results.

Showroomprivé engaged a team of thirty employees to write its product sheets, at a rate of five minutes per sheet, for a catalogue that renews with each sale. Galadrim developed a web application that generates descriptions for an entire sale, based on product characteristics as well as their photographs. Teams now proofread and edit instead of writing, and put a sale online with a consistent level of detail from one reference to another.

A major player in insurance brokerage, Odealim receives tens of thousands of customer requests via email each month, which are challenging to process at scale while maintaining service quality. Galadrim designed a system of AI agents that classifies requests, searches for useful information in contracts and files, and generates response proposals. Each response is validated by the business teams via a dedicated interface, with human oversight for sensitive matters.