Knowledge accuracy
Every answer needed to be grounded in his real writing, not approximations that lose the nuance of his actual views.
AI Application DevelopmentGenerative AIWeb App DevelopmentUI/UX Design
FabriceAI
AI-Powered Knowledge Assistant
ProjectOverview
Fabrice Grinda has spent decades building, investing, and writing. Hundreds of blog posts, years of investment thinking, and a philosophy about markets and opportunity that very few people have had the chance to sit down and really absorb. The knowledge was all there, but it was scattered, and accessing it meant hours of reading with no guarantee you would find exactly what you were looking for.
That is where FabriceAI begins. We built a real-time conversational platform where anyone can have a natural voice or text conversation with an AI that thinks, responds, and reasons the way Fabrice does. Ask it about market sizing, a specific industry, or the reasoning behind a past investment, and you get an answer grounded in his actual words, his actual thinking, with the sources right there to follow if you want to go deeper.
TheChallenge
Fabrice's body of work is vast and deeply personal. Making it genuinely useful through AI required solving several meaningful problems.
Document intelligence
Users needed to share PDFs, pitch decks, and documents and have the AI reason over them during the conversation.
Natural conversation
The interaction needed to feel like talking to someone, with a fluid, low-latency experience, not a stilted query box.
Continuity across sessions
Returning users should be able to continue a previous conversation without starting over every time.
The Solution
We designed and built FabriceAI as a purpose-built conversational AI platform, combining real-time voice technology, retrieval-augmented generation, and document intelligence.
At the heart of the platform is a curated vector database built from Fabrice’s entire blog archive. Every response is grounded in this source material, with cited blog posts surfaced alongside each answer, so users always know exactly where the information comes from.
For voice, we integrated OpenAI’s Realtime API with semantic voice activity detection, enabling natural, low-latency conversations. Users can upload PDFs and decks, which the system reads with OCR and vision models. Conversation history is stored in MongoDB so every session picks up where the last one left off, and the platform detects the user’s language automatically.
KeyFeatures
A purpose-built conversational AI platform, combining real-time voice, retrieval-augmented generation, and document intelligence.
01
Real-time voice
Real-time voice
and text chat
Natural, low-latency voice conversations powered by OpenAI’s Realtime API with semantic voice activity detection.
02
Retrieval-augmented
Retrieval-augmented
generation
Every response is grounded in a curated vector database built from Fabrice’s entire blog archive.
03
Source citations
Cited blog posts are surfaced alongside each answer, so you always know exactly where the information comes from.
04
Document
Document
intelligence
Upload PDFs and pitch decks, which the system reads with OCR and vision models to reason over during the conversation.
05
Persistent
Persistent
session memory
Conversation history is stored in MongoDB, so every session picks up right where the last one left off.
TechStack
Next.js, real-time voice, and retrieval-augmented generation, built to keep every answer grounded and fast.
Next.js 19
React
OpenAI Realtime APIOpenAI Agents SDK
MongoDB
AWS S3
Tailwind CSS
TypeScript
Business Impact
Ready to make expertiseinstantly accessible?
See how FabriceAI can turn a deep knowledge archive into grounded, cited conversations available whenever users need an answer.
Get started