Team
404 Brain Not Found
This team is not accepting new members.
Project Concept
No description has been added yet.
Entry
Status: Submitted
Last saved: November 15 at 6:03 PM CET
Team Roster This team is not accepting new members.
Message board not available for this team yet.
Mallet Romain Team Lead
Student at ECE
For my part, I focused on everything related to the core intelligence of the system. I built the entire pipeline that powers the AI agent: the reasoning layer, the call-handling logic, the message parsing, and all the integrations between Mistral, ElevenLabs, and the backend. I tested multiple Mistral models, compared their speed and stability, and ultimately chose Mistral Medium 3.1 for its balance of accuracy and real-time performance during phone conversations.
I also worked on connecting the voice agent with ElevenLabs, managing the end-to-end audio pipeline, experimenting with different webhook strategies, and ensuring the agent could listen, respond, and act instantly without lag. My goal was to make the AI feel like a real person who can understand the caller perfectly, extract the right information, and complete the reservation on its own.
To maximize our chances of success, we organized the project with a dual-path strategy: at every stage, there were two different technical approaches possible. My teammate and I each explored one path in parallel — I took one route, he took the other. This allowed us to validate ideas twice as fast, identify the most reliable solution, and always have a functional fallback. This approach is what let us build a fully working end-to-end product in such a short time.
Overall, my contribution was to make the AI agent smart, fast, and autonomous, and to ensure that the voice experience, from reasoning to audio streaming, worked flawlessly for the final demo.
Salut
Slut Salut
I'm currently working on an autonomous voice-based AI agent capable of handling real phone calls end-to-end, without human intervention.
The goal is to build a system that can receive a call, transcribe speech in real time, understand the request, take actions, update a database or a Google Sheet, and answer naturally using a local LLM.
The agent is designed as a mini one-person company: it can manage bookings, customer support, or restaurant orders autonomously.
The architecture integrates a r
Baptiste SEITZ
student at ECE
For my part, I focused on everything related to the product experience, the UI, and the reliability of the system. I built the entire front-end using Lovable, designing a clean, simple, and intuitive dashboard where companies can create their phone number, view their call history, access customer profiles, and see all the information extracted by the AI. My goal was to make the interface feel like a real, production-ready SaaS, even within the short timeframe of the hackathon.
While my teammate worked on the RAG pipeline, the agent logic, and the integrations with Mistral and ElevenLabs, I explored all alternative technical paths in parallel. For every step of the project, we deliberately split the work into two different approaches — he tested one, and I tested the other. This allowed us to validate ideas extremely quickly and always have a fully functional fallback. I also handled the testing and debugging of the voice agent flow, experimented with different webhook configurations, and ensured the demo data and the call summaries could be displayed cleanly in the UI.
In short, my contribution was to guarantee that the product was usable, polished, visually coherent, and that every part of the system was stable enough for a seamless demo. Our dual-path strategy is what allowed us to move twice as fast and deliver a complete end-to-end experience today.
Baptiste Seitz is an early-career engineering student (Master’s Student in Data & AI at ECE Paris) currently completing internships and practical engineering work at ARGEDIS. He has a Diplôme d'ingénieur in progress from ECE, participated in an exchange at Baruch College (courses including introduction to AI, Intelligent Robotics and Financial Management), and holds a scientific Baccalauréat from Gerson. At ARGEDIS he worked on Power BI reporting, data centralization tooling, information system exploration and cybersecurity tasks (firewall, phishing mitigation, Microsoft Defender), and developed Python scripts to extract insights. Prior short-term roles include facilities/assistant roles at Electre and merchandising at Carrefour. Academic projects show hands-on experience in embedded systems, FPGA design, electronics, computer vision (YOLO/OpenCV), robotics (Raspberry Pi), C/Java/Python development and ML/vision applications, positioning him as an early-stage technical candidate for data engineering, computer vision, or embedded/IoT internships.