Menu
Fabrice Grinda is one of the world's most active angel investors, with decades of thinking, writing, and investing distilled across hundreds of blog posts and insights. The challenge was not simply preserving all of that knowledge — it was making it accessible in a way that felt alive, personal, and genuinely useful to anyone curious about his perspective.
We built FabriceAI: a real-time conversational platform where users can have natural voice and text conversations with an AI that speaks with Fabrice's knowledge, philosophy, and point of view. Whether you want to understand how he thinks about market sizing, his take on a specific industry, or the reasoning behind a past investment, FabriceAI gives you a direct window into his world.
—
An AI that simply generates responses would lose the nuance and specificity of Fabrice's actual views. Every answer needed to be grounded in his real writing, not approximations.
Users expect a fluid, low-latency experience — not the stilted back-and-forth typical of many AI tools. The interaction needed to feel like talking to someone, not querying a search engine.
Beyond blog content, users needed to share PDFs, pitch decks, and documents, and have the AI reason over them intelligently during the conversation.
A meaningful knowledge tool needs memory. Users returning to continue a previous conversation should not have to start over every time. —
We designed and built FabriceAI as a purpose-built conversational AI platform, combining the latest in 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 the AI generates is grounded in this source material, with cited blog posts surfaced alongside each answer — similar to ChatGPT's citation pane. Users always know exactly where the information is coming from.
For voice interaction, we integrated OpenAI's Realtime API with semantic voice activity detection, enabling natural, low-latency conversations that feel genuinely responsive. Users can also upload PDFs, pitch decks, and documents, and the system uses OCR and vision models to extract and reason over that content in real time.
Conversation history is stored persistently in MongoDB, so every session picks up where the last one left off. The platform also detects the user's language automatically, responding in their preferred language without any manual configuration.
—
Built on OpenAI's Realtime API with semantic voice activity detection, enabling natural, low-latency conversations across both voice and text.
Every AI response is grounded in a curated vector database of Fabrice's blog archive, ensuring answers are accurate, relevant, and always traceable.
Linked blog posts are surfaced alongside each AI response, giving users full transparency into where the information originates.
Users can upload PDFs, pitch decks, and documents. The system uses OCR and vision models to extract and reason over the content during the conversation.
Conversation history is stored in MongoDB, enabling continuity across sessions so users can pick up naturally where they left off.
Automatic language detection with multilingual response capability, serving users across different regions without any extra configuration. —
FabriceAI turns a vast archive of investment thinking into an always-on, always-accurate knowledge partner. Whether it is a founder preparing to pitch, an investor doing research, or a reader exploring a new idea, the platform makes Fabrice's perspective accessible in a way that scales far beyond what any blog or interview ever could.
—
*End of Case Study — FabriceAI*
Tell us what you’re building. We’ll come back with a clear, fixed-scope plan to ship it.
