Daynis OlmanAI, Cloud & Digital Platform Leader

Conversational Operations Intelligence

A conversational analytics and operational-intelligence platform that turns complex safety and event data into trusted dashboards, inspectable queries and grounded natural-language answers.

From operational safety data to trusted answers.

  1. 1. Operational safety data
  2. 2. Secure, incremental ingestion
  3. 3. BigQuery and governed intelligence
  4. 4. Trusted dashboard and conversational answer
Evidence: Confidential case studyAI, data and Google Cloud· Seeing Machines· Internal identifiers and datasets withheld
Role
Architecture and technical leadership
Serves
Operational decision-makers and analysts who rely on safety and event data.
Core technologies
BigQuery · Gemini · Firebase · React

Executive overview

The problem. Operational safety and event data was rich but complex. Decision-makers needed answers without waiting on analysts, yet any AI layer had to stay traceable and trustworthy rather than producing unverifiable numbers.

Who it serves. Operational decision-makers and analysts who rely on safety and event data.

What Daynis did. Architecture and technical leadership. A conversational analytics and operational-intelligence platform that turns complex safety and event data into trusted dashboards, inspectable queries and grounded natural-language answers.

Why it matters. Makes operational information accessible to decision-makers while preserving traceability, governance and analytical trust. Common questions follow deterministic paths; AI assistance is bounded by validation and read-only controls.

Business architecture and impact

Capabilities created

  • Self-service dashboards over operational safety and event data
  • Natural-language questions answered from governed analytical data
  • Transparent SQL behind every dashboard result
  • Conversation history so lines of inquiry can be resumed and reviewed

Makes operational information accessible to decision-makers while preserving traceability, governance and analytical trust. Common questions follow deterministic paths; AI assistance is bounded by validation and read-only controls.

System architecture

  1. Decision-makers & analysts + Safety & event sources

    Where requests and source data originate.

  2. React + Firebase app

    The interface people use day to day.

  3. Cloud Functions orchestration

    Orchestrates requests and enforces business rules.

  4. Gemini NL-to-SQL

    Models applied under defined, reviewable constraints.

  5. BigQuery analytical store

    The governed system of record.

Governance: Query validation, read-only access
Delivery: Secure incremental ingestion
Users interact through a React and Firebase application. Cloud Functions orchestrate requests: common questions route to deterministic analytical queries, while open questions pass through Gemini-assisted natural-language-to-SQL, a validation step and read-only execution against BigQuery. Results are summarised from the returned data only, and the generated SQL remains visible beside every answer. Data arrives through secure incremental ingestion.

Key flows

  • Decision-makers & analysts to React + Firebase app
  • React + Firebase app to Cloud Functions orchestration
  • Cloud Functions orchestration to Deterministic query paths (common questions)
  • Cloud Functions orchestration to Gemini NL-to-SQL (open questions)
  • Gemini NL-to-SQL to Query validation, read-only access

Technical depth

Highlights

  • BigQuery analytical platform with secure incremental ingestion
  • React and Firebase experience backed by Cloud Functions
  • Gemini-assisted natural-language-to-SQL with query validation
  • Deterministic analytical paths for common questions
  • Read-only query controls and grounded summarisation

Architecture decisions

Deterministic first, AI second

Frequently asked questions run through known, tested query paths. Generative SQL is reserved for questions those paths do not cover.

Show the SQL

Every dashboard result exposes the query behind it, so analysts can inspect and trust what they see.

Grounded answers only

Natural-language summaries are generated from returned results, not from model knowledge.

Incremental ingestion

Data lands securely and incrementally rather than through bulk reloads, keeping the analytical store current.

Security and governance

  • Generated queries are validated before execution
  • Read-only controls prevent any data modification from the AI path
  • Answers are traceable to the query and data that produced them

What I led

  • Architecture ownership across ingestion, analytics and AI layers
  • Technical direction on where AI is appropriate and where deterministic logic is required
  • Stakeholder engagement translating operational questions into analytical capability

Evidence and links

Evidence: Confidential case studyShared at a public-safe level.

Related speaking topic

Building something similar?

Talk to Daynis about ai, data and google cloud architecture and delivery.

Start a conversation