# Rizlum AI > On\-premise AI agents for insurance and banking, built on sovereign Small Language Models\. ## At a glance | Field | Value | | --- | --- | | Category | Claims Technology | | Type | Solutions Provider | | Region | EMEA | | Headquarters | Paris, France | | Founded | 2024 | | Stage | Startup | | Employees | 11-50 | | Funding | Bootstrapped | ## Overview Rizlum AI is an AI agent platform for insurance carriers and mutual health insurers, offering document processing, voice, email, and fraud detection automation built on fine-tuned Small Language Models (SLMs). Founded in 2024 at Station F in Paris by three Ecole Polytechnique de Hanoi graduates -- Theo Pham, Quang-Linh Le, and Dang-Chuan Nguyen -- the company targets regulated sectors in France where data sovereignty and on-premise deployment are hard requirements. The platform deploys SLMs with 2-7 billion parameters rather than large language models, allowing full operation behind the client firewall with no external API calls. Three studios -- AgentStudio, WorkflowStudio, and TrainingStudio -- allow insurers to configure pre-built agents, build no-code automation pipelines, and fine-tune models on proprietary data within a 48-hour retraining cycle. Deployment options include on-premise, private cloud, and SaaS. Published case studies from French mutual insurers report 185,000 member integrations in under two months, 94% accuracy on handwritten document extraction, and a 700% ROI on a ChatAgent deployment. The company is a laureate of the French Government AMI IA program and has joined accelerator programs run by Google for Startups, OVHcloud, Microsoft for Startups, and Bpifrance. ## Products & services ### AgentStudio Interface for building and configuring AI agents from a pre-built catalog. Business teams can compose process automations using existing agent building blocks without writing code. **Key features:** - Pre\-built agent catalog - Configurable agent parameters - No\-code composition interface *Target users:* Insurance operations and IT teams ### WorkflowStudio No-code workflow editor for orchestrating modules, agents, and API calls. Connects Rizlum agents with existing enterprise systems including GED, CRM, and back-office platforms. **Key features:** - No\-code pipeline design - Multi\-step orchestration of agents and API calls - Integration with existing enterprise IT infrastructure *Target users:* Operations managers, business analysts ### TrainingStudio Fine-tuning platform allowing clients to retrain SLMs on proprietary data. Features a 48-hour retraining cycle with all data remaining within the enterprise perimeter. **Key features:** - Custom SLM fine\-tuning on client data - 48\-hour retraining cycle - Zero data exfiltration *Target users:* IT and data teams at insurance companies ### DocInsight AI document extraction agent supporting over 10 document types including medical invoices, work stoppages, handwritten documents, identity documents, prescriptions, Social Security attestations, bank account documents, and family records. **Key features:** - Greater than 94% accuracy on handwritten documents - Supports 10\+ document types - On\-premise processing with no third\-party data sharing *Target users:* Claims processors, member services teams ### VoiceAgent Fully on-premise voice agent using a speech-to-text to SLM to text-to-speech pipeline with under 800ms latency. No data is transmitted to third-party voice APIs. **Key features:** - Under 800ms response latency - 100% on\-premise, no third\-party voice APIs - Full STT\-SLM\-TTS pipeline *Target users:* Contact center operators ### CallInsight Real-time call transcription with greater than 97% accuracy and automatic call summarization for contact center quality assurance, compliance, and coaching. **Key features:** - Greater than 97% transcription accuracy - Automatic call summarization - Quality assurance and compliance monitoring *Target users:* Contact center supervisors, compliance teams ### ChatAgent Conversational assistant using Retrieval-Augmented Generation (RAG) on client documentation. Deployed in omnichannel contexts to handle level-1 support queries. **Key features:** - RAG on client\-specific documentation - Omnichannel deployment \(web, messaging\) - Reported 60\-70% reduction in level\-1 support calls *Target users:* Policyholder services, member support teams ### EmailInsight Email classification and intelligent attachment categorization with greater than 90% automation rate. Handles routing, prioritization, and content extraction from inbound email flows. **Key features:** - Greater than 90% automation rate - Intelligent routing and prioritization - Attachment categorization *Target users:* Back\-office operations teams ### FraudDetection Documentary fraud detection using metadata analysis and AI-generated content identification, including pixel-level image manipulation detection in claims and onboarding documents. **Key features:** - Pixel\-level image manipulation detection - AI\-generated document identification - Metadata analysis *Target users:* Claims investigators, underwriting teams ## Category & focus - Subcategories: AI Agents, Document Processing, Fraud Detection, Customer Engagement, Voice & Speech - Insurance verticals: Health, P&C Commercial, P&C Personal - Target customers: Carriers ## Notable customers - Unnamed French mutual insurer \-\- 185,000 member integration in under 2 months using DocInsight - Unnamed French mutual insurer \-\- 50,000 annual invoice automation at 94% success rate - Unnamed French mutual insurer \-\- ChatAgent deployment reporting 700% ROI - Unnamed French insurer \-\- Claims processing reduced from 3\-6 days to 5\-10 minutes ## Links - Website: - Directory profile: - Linkedin: --- *Last updated: 2026\-06\-08*