Menu
A general-purpose AI would hallucinate or generalise. Every answer needed to come from Fabrice’s actual writing, not a model’s best guess at what an investor might say. Accuracy was non-negotiable.
Nobody wants to interact with an AI that sounds like a search engine. The experience needed to feel warm, fluid, and low-latency, whether the user was typing or speaking.
Users often arrive with pitch decks, PDFs, and research notes they want to discuss. The AI needed to actually read and reason over those documents in the conversation, not just acknowledge them.
A knowledge tool that forgets everything the moment you close the tab is not really useful. Users needed their conversations to persist and continue naturally across sessions.
Our development team designed and deployed FabriceAI, a robust AI-powered web application tailored to present Fabrice’s viewpoints through a seamless, engaging user interface.
We started by building a curated vector database from Fabrice’s entire blog archive. This is the memory at the heart of FabriceAI.
Every time a user asks something, the system searches this database and grounds its response in real source material, surfacing the relevant blog posts alongside every answer so the user can always trace where an idea came from.
For voice, we integrated OpenAI’s Realtime API with semantic voice activity detection, which means the AI knows when you have finished speaking and responds naturally, without awkward pauses or interruptions.
For documents, we built an OCR and vision pipeline that lets the AI extract and reason over PDFs and pitch decks in real time, mid-conversation.
And because continuity matters, we store every conversation in MongoDB so users can return days later and continue exactly where they left off, in whichever language they prefer.
Natural, low-latency conversations powered by OpenAI’s Realtime API, with semantic voice activity detection that makes voice interactions feel genuinely human.
Every response is anchored in Fabrice’s actual blog archive through a curated vector database, so answers are accurate, traceable, and grounded in real thinking.
The relevant blog posts surface alongside every AI response, giving users the transparency to follow up, verify, and explore further whenever they want.
Users can upload PDFs, pitch decks, and research documents. The system uses OCR and vision models to read and reason over them during the conversation, in real time.
Conversation history lives in MongoDB across sessions, so every returning user picks up naturally from where they left off, with no need to repeat context.
The platform detects the user’s language automatically and responds in kind, making Fabrice’s thinking accessible to a global audience without any configuration.
Explore how the innovative AI solution (AIDA: Your AI-Powered Ally in Legal Dispute Resolution)
Discover how the web application for Zkrillafi transforms every bill payment into a credit-boosting
opportunity!
