Grundl Leadership Institut

GRUNDL.ai

GRUNDL.ai makes the knowledge of Boris Grundl available as a chat and AI real-time conversation.

GRUNDL.ai is the custom expert GPT of the Grundl Leadership Institute and makes the knowledge of founder Boris Grundl available exactly when it is needed: in the moment of a concrete leadership question.

Users enter their questions directly in the chat and immediately receive well-founded answers that are based on the content of the existing learning platform. Every answer contains direct references to matching lessons or book passages, so users can dive deeper right away. In addition, a realtime voice app was built that enables a live conversation with the cloned voice of Boris Grundl.

As the founder, it matters to me that my content truly helps people in their everyday life. GRUNDL.ai makes access to this knowledge significantly easier."
Boris Grundl

How it works

Wissen schwer auffindbar?6 Bücher9.000 Min.

Starting point and objective

Where it began and where it should go

Starting point
  • The Grundl Institute runs a digital learning platform with approximately 9,000 minutes of video content and 6 books of about 200 pages each.
  • A large share of the knowledge exists as video, which makes it hard for users to quickly find a specific passage or relevant content.
  • An initial in-house solution for search and answers was already in place, but was technically not robust enough (issues around performance and streaming; partly wrong language).
Objective
  • Phase 1: Users should be able to go to the right spot (book chapter/lesson/video) directly from a chat inside the learning platform and jump into the corresponding content.
  • Phase 2: In addition to chat, a realtime conversation should become possible (calling with Boris' cloned voice) to make the interaction more natural. (approx. 30 minutes of standard use)
LoginFrage stellenRetrievalLLM AntwortAntwort + LinksVoice-App

Process chain

From login to answer, in chat and voice

GRUNDL.ai Chat (RAG-based)
  1. 1.Login and entry: user signs in, sees usage hints and example questions
  2. 2.Question in the chat: user asks a question or follow-up in the context of the conversation
  3. 3.Retrieval: the system finds matching content from the knowledge base
  4. 4.Answer generation: the LLM formulates an answer based on the retrieved sources
  5. 5.Transparency and linking: the output contains the answer and relevant lessons/sources; users can jump there
Voice app (Leading Simple app)
  1. 1.User starts a voice session and speaks into the microphone
  2. 2.Voice activity detection and transcription (realtime): text input for the AI system
  3. 3.RAG/LLM generates an answer. Voice cloning streams audio back
  4. 4.End of session: transcript
  5. 5.Perspective: integration as a menu item in the online learning world („call Boris Grundl")

Architecture highlights

Core building blocks

  • Heterogeneous knowledge base made usable: books and videos (via transcripts) ingested and prepared so they can be reliably found and referenced.
  • Realtime voice pipeline with a focus on latency. Target image „very fast". Identified lever: embeddings/roundtrips, with on-device embeddings as a future option.
  • Clean interrupt and turn taking: users can interrupt the output by speaking in between. Technically demanding handling of blocking/streaming.
  • Video ingestion for video search: transcription as a foundation so video content becomes semantically searchable and can be referenced in chat.
  • Integration into the existing learning platform: chat system as part of the learning world (instead of a separate tool), including user guidance.