Skip to content

2026 · Analyse

Turning Company Recordings into Searchable Knowledge

Delivered a cross-media RAG & NLP tool for Synchrone — OCR, speech-to-text, chunking, embeddings, vector indexing, SSO authentication and a chatbot with adjustable temperature to query company knowledge across any media format.

DeliveredRAGNLPOCREmbeddingsChatbotSSODataiku

Interactive demo

Verba, the knowledge base we built

Ingest recordings, then question them. Every answer cites the file and the exact minute it comes from, and Verba says “I don't know” rather than inventing. Pick a project, add a sample recording or drop your own text file, and ask away.

Simulated in your browser · illustrative data

Loading the demo…

The challenge

Synchrone's expertise is captured across meetings, training sessions, documents and engineers' recordings, making it hard to search at scale. Carried out during a three-week Business Deep Dive, the goal was to build a unified RAG/NLP platform capable of ingesting any media format and making it conversational.

The question

How can heterogeneous company media (video, audio and documents) be turned into a secure, searchable knowledge base that employees can query through an AI chatbot?

What I did

Built an end-to-end multimodal ingestion and RAG pipeline: speech-to-text for audio/video and OCR for documents, semantic chunking, embedding generation and vector indexing. Exposed the knowledge base through an interactive chatbot featuring adjustable generation temperature and secured access via Single Sign-On (SSO).

Backstage · Data sources

Multi-format company media starting from a kickoff sample of seven videos and designed to ingest audio recordings, videos and text/PDF documents at scale across thousands of hours of content.

The result

Delivered a working RAG/NLP tool handling any media format, complete with SSO authentication and a customizable chatbot interface (including temperature control) for exploring enterprise knowledge.