Daynis OlmanAI, Cloud & Digital Platform Leader

Daynis Notes: Grounded Knowledge Intelligence

A personal knowledge and meeting-intelligence system combining notes, tasks, briefings, transcription and grounded answers with source-aware citations.

Every answer traces back to a source.

  1. 1. Notes and transcripts
  2. 2. Embeddings and indexing
  3. 3. Retrieval and grounding
  4. 4. Cited answer
Evidence: Public productRAG, AI and mobile
Role
Product design, architecture and engineering
Serves
Professionals managing meetings, notes and tasks.
Core technologies
Flutter · Vertex AI · Gemini · Firebase

Executive overview

The problem. Most “chat with your notes” tools are a chatbot over a vector search. Answers are often unverifiable and quietly wrong. Credible retrieval needs quality controls the user can see.

Who it serves. Professionals managing meetings, notes and tasks.

What Daynis did. Product design, architecture and engineering. A personal knowledge and meeting-intelligence system combining notes, tasks, briefings, transcription and grounded answers with source-aware citations.

Why it matters. Demonstrates that credible RAG requires retrieval quality, source visibility, validation, recovery and product design. It is not simply a chatbot interface.

Business architecture and impact

Capabilities created

  • Notes, tasks and briefings in one place
  • On-device meeting transcription
  • Answers grounded in the user's own content, with citations
  • Source cards showing where each answer came from

Demonstrates that credible RAG requires retrieval quality, source visibility, validation, recovery and product design. It is not simply a chatbot interface.

System architecture

  1. User

    Where requests and source data originate.

  2. Flutter app, on-device transcription

    The interface people use day to day.

  3. Query & timeframe routing + Context assembly

    Orchestrates requests and enforces business rules.

  4. Vertex embeddings

    Models applied under defined, reviewable constraints.

  5. Vector retrieval (Firebase)

    The governed system of record.

Governance: Confidence scoring, response validation
Delivery: Embedding lifecycle & recovery
A Flutter app captures notes and transcribes meetings on-device. Content is embedded with Vertex and stored for vector retrieval in Firebase. Questions are routed by intent and timeframe, retrieved results are diversity-filtered and confidence-scored, and assembled context is passed to Gemini. Responses are validated against a structure and shown with source cards.

Key flows

  • User to Flutter app, on-device transcription
  • Flutter app, on-device transcription to Vertex embeddings
  • Vertex embeddings to Vector retrieval (Firebase)
  • Flutter app, on-device transcription to Query & timeframe routing
  • Query & timeframe routing to Vector retrieval (Firebase)

Technical depth

Highlights

  • Vertex embeddings with vector retrieval
  • Query and timeframe routing, diversity filtering and confidence scoring
  • Context assembly and grounded Gemini generation
  • Structured response validation
  • Embedding lifecycle and recovery controls

Architecture decisions

Route before retrieving

Queries are classified by intent and timeframe so retrieval looks in the right place.

Diversity and confidence

Results are filtered for diversity and scored for confidence before they reach the model.

Citations are product, not decoration

Source cards make every answer checkable by the user.

Plan for recovery

Embeddings have a managed lifecycle so the index can be repaired or rebuilt.

Security and governance

  • Transcription happens on-device
  • Structured validation of model responses
  • Answers limited to the user's own sources

What I led

  • End-to-end product design and architecture
  • Engineering standards for retrieval quality and evaluation

Evidence and links

Evidence: Public productPublicly available product.

Related speaking topic

Building something similar?

Talk to Daynis about rag, ai and mobile architecture and delivery.

Start a conversation