September 2025
Swizy – RAG AI
From PDF Chaos to Instant AI Assistance: Automated Employee Benefits Management
Learn more
Project Introduction
Swizy is a French employee benefits management platform designed to streamline corporate perks for modern companies. Managing an extensive catalog of offerings-ranging from gym memberships to cultural grants-presented a major operational bottleneck.
Every vendor delivered their own documentation in distinct PDF formats, detailing complex eligibility criteria such as age limits, tenure, contract types, and deadlines.
Every vendor delivered their own documentation in distinct PDF formats, detailing complex eligibility criteria such as age limits, tenure, contract types, and deadlines.
With dozens of providers issuing unstructured files filled with text, tables, and diagrams, traditional search methods were ineffective. Finding simple answers about a specific benefit took employees anywhere from several hours to a few days.
Swizy needed an intelligent, modern solution capable of instantly processing complex documents and presenting accurate answers through a natural conversational interface.
Swizy needed an intelligent, modern solution capable of instantly processing complex documents and presenting accurate answers through a natural conversational interface.
Technology Stack
Cloud Infrastructure
Amazon Web Services
Core Language
Python
AI Framework
LangChain (RAG pipeline orchestration)
Database
PostgreSQL
Project Scope & Core Features
As a full-spectrum AI partner, ITSharkz engineered a complete end-to-end platform:
- RAG Knowledge Engine: Automated ingestion and processing of multi-format vendor PDF documentation.
- Table & Diagram Resolution: Intelligent layout parsing to accurately read structured grids, schedules, and pricing diagrams.
- Browser Chat Application: A clean, responsive frontend tailored for seamless interaction with the AI assistant.
- Embeddable Architecture: Designed with future growth in mind, enabling easy integration directly into Swizy’s client portals.
- Multi-Tenant Data Isolation: Enables Swizy to serve multiple B2B clients on a single platform while strictly isolating each company’s document collection into dedicated database namespaces.
Business Challenge:
- Unstructured Knowledge Base: Inconsistent vendor PDFs made traditional cataloging and keywords useless.
- Time-Consuming Search Process: Users lost hours navigating dense terms and conditions.
- Risky Architectural Strategy: The initial plan was to host open-source AI models on local servers to avoid API vendor fees.
- Enterprise Data Privacy (Multitenancy): Standard RAG architectures rely on a single monolithic database. Swizy required a multi-client platform where each corporate customer has its own confidential benefit files.
ITSharkz Strategic Partnering & Solution:
- RAG Architecture with Advanced Parsing: We engineered a data pipeline that ingests complex PDFs into a vector database. To maintain 95%+ accuracy across complex diagrams and tables, we incorporated advanced document parsing.
- Intelligent AI Assistant: We designed and launched a browser-based web application with an intuitive chat UI. The assistant provides instant responses, contact details, and policy terms within seconds.
- Proactive Technical Guidance: We provided strategic consulting, demonstrating that local hosting would result in exorbitant infrastructure costs, heavy maintenance overhead, and production instability. We guided Swizy toward an optimized cloud strategy.
- Multi-Tenant RAG Architecture: We engineered a dynamic routing mechanism that directs RAG queries to isolated database namespaces/instances per enterprise client. A single engine seamlessly serves multiple clients with zero risk of cross-company data leakage.
Features & Capabilities
Context-Aware RAG Engine:
Natural language processing that extracts accurate answers directly from multi-vendor PDF documentation.
Visual Table & Layout Parsing
High-precision visual parsing capable of reading complex tables, price grids, and diagrams without layout distortion.
Dedicated Web Chat UI
A clean, responsive browser-based chat interface built specifically for smooth conversational AI interaction.
Automated Eligibility Verification
Intelligent rule-matching based on employee attributes (age, tenure, contract type, deadlines).
Scalable AWS Infrastructure
Cloud-native deployment ensuring rapid, multi-second response times under varying loads.
Testimonials
What Our Partners Say
Goals & Business Outcomes
- Instant Access to Information: Reduced retrieval times from hours/days to just a few seconds.
- Exceptional Accuracy (95%+): Advanced visual parsing prevented layout distortion, ensuring accurate data extraction.
- Long-Term Cost Efficiency: By guiding Swizy away from self-hosted models, we saved them significant capital expense and eliminated production maintenance nightmares.
Conclusion
The Swizy case study highlights ITSharkz’s positioning as a strategic AI automation partner. Moving far beyond traditional execution, we acted as trusted advisors by guiding the client away from an inefficient local model hosting strategy, saving them from severe maintenance overhead and financial waste.
By engineering a robust RAG architecture paired with visual document parsing, we transformed chaotic, unstructured PDF data into a accurate, real-time AI assistant.
This project underscores our ability to solve complex data challenges and integrate tailored AI solutions that drive real operational value.
By engineering a robust RAG architecture paired with visual document parsing, we transformed chaotic, unstructured PDF data into a accurate, real-time AI assistant.
This project underscores our ability to solve complex data challenges and integrate tailored AI solutions that drive real operational value.
Watch our client’s video reviews
Nicolas Baudelot and Nicolas Massaviol – CEO and CTO at Medicalib
"With ITSharkz, we found a trusted partner, so we will continue working with them – and hopefully for a long time."
Tasks:
- Integrated government systems, implemented regulatory logic, and created a UI library.
Results:
- Integrated OpenID Connect, developed a quotation engine, and built a design system.
Carole Juge-Llewellyn – CEO and Founder of JOONE
"They really got involved in understanding what we do, how we do it, and were proactive. That’s something I’ve never seen elsewhere."
Tasks:
- Developed new subscription checkout, implemented client-facing management system, and migrated thousands of subscriptions.
Results:
- Launched Newcharge checkout, built a customer dashboard for subscription management, and successfully migrated subscriptions.
Georges Fornay and Franck Véga – CEO and CTO at Qobuz
"It was the right compromise for me: very skilled people, competitive rates, and a nearby location."
Tasks:
- Optimized backend, modernized desktop app, integrated cross-platform features (Qobuz Connect).
Results:
- Improved device support & app performance, boosted user retention, cut costs, and future-proofed systems.
Rémi Prévost – CTO at Happydemics
"ITSharkz isn't just in a headhunter position that gives you a profile and then you're on your own. They are truly in a process of accompaniment from the beginning to the end."
Tasks:
- Curating a dedicated squad of senior Ruby on Rails experts to solve the high-level talent gap for a French scale-up.
- Filling the positions with product-first mindset engineers, ensuring long-term alignment with internal engineering standards.
Results:
- Successfully integrated 4 senior developers into the product roadmap, eliminating "mercenary" turnover through a culture of ownership.
- Accelerated feature delivery while maintaining full CIR (R&D Tax Credit) compliance, providing a high-ROI strategic advantage.
Matthieu Derancourt – Medical Advisor at Pauchet Santé
"ITSharkz delivers a product that is truly near-complete, so we don't waste any time. It isn't designed by engineers, but really by the end-user."
Tasks:
- Industrializing complex hospital processes that had reached the limits of standard Office tools.
- Bridging the gap between medical strategy and technical execution to ensure software perfectly matches the clinical reality.
Results:
- Successfully deployed 3 mission-critical platforms with 90% operational readiness at the first delivery stage.
- Streamlined the development lifecycle using a Design-First approach, resulting in high user adoption.