Find a time with me

For Andra and the Airalo data team

Every path to your data runs through Lightdash.

Dashboards, self-service, Lightdash agents, Claude, Slack bots: five ways in, one governance layer behind all of them. What we build for each, and how we run the work.

Bijan Soltani

Bijan Soltani · Founder & Managing Director, Gemma Analytics

Hi Andra,

Good talking through where Airalo's data setup is heading. Self-service in Lightdash, a semantic layer that can carry more weight, and AI enablement on top of Lightdash, with the org-wide BigQuery access line you drew. All three are things we work on every week.

This page makes that concrete: what data access looks like with Lightdash in the middle, and how we'd run the engagement, measured against what you told me matters most.

The usage side

How your teams work with Lightdash

Whoever is asking, and whichever tool they ask from, the request goes through Lightdash's semantic layer and its permissions. No direct BigQuery access, for anyone.

Dashboards

Ready-made, per team.

Self-service

Explore and build directly in Lightdash.

Lightdash agents

Ask questions or sketch charts inside Lightdash.

Claude Code / Codex

Via Lightdash MCP: ask questions, or build dashboards through the API.

Slack bots

Ask in Slack, get the answer from Lightdash.

Lightdash Semantic layer · Metrics + dimensions · Row filters compiled into the SQL
BigQuery + dbt models
What we do for you

Implementation and support

Every path above is something we can build or set up for you.

Dashboards and self-service

Migrate legacy BI to Lightdash

Looker and QuickSight dashboards ported over, semantic layer first.

Dashboards and self-service

Build out metrics and dimensions

New dimensional modeling in dbt, so more of the org's questions have an answer in Lightdash. Greenfield work, not a port of what already exists.

Lightdash agents

Configure one agent per team or function

Each with the context its domain needs, so "a year" means the same thing to the agent as to the team using it.

Claude Code / Codex

Set up Lightdash MCP

For Claude Code and Codex alike: only ever through the governance layer, inheriting the user's access, so someone sees in either tool exactly what they would see in Lightdash.

Slack bots

Build the data-analyst bot

On Lightdash MCP, in the channels people already use. When it can't answer, it opens a ticket for your BI team.

Every path

Make it reliable

Golden records and evaluation sets with expected answers, so you can measure agent quality.

The standard

What makes a semantic layer agent-ready

You said there is no clear recipe for this yet. Here is ours: the checklist we run a semantic layer through before agents get access to it.

  1. One definition per metric

    Revenue means one thing, and everything built on top of it inherits that meaning. If two definitions survive, the agent picks one, and whoever reads the answer has no way of knowing which.

  2. Descriptions the agent can search

    Tables, fields, metrics and dimensions described well enough that the model finds the right one instead of guessing from its name. This is most of the work.

  3. No direct SQL

    Agents query through metrics and dimensions. Give them raw SQL and they start running in every direction, with none of your permissions attached.

  4. Somewhere to escalate

    The agent needs to be able to say it cannot find something and turn that into a ticket for your BI team. Otherwise the gap never reaches anyone who could fill it.

  5. Golden records

    A set of questions with known answers, so you can check the agents still work after someone changes a model.

Process

How an engagement with us runs

One named lead, one named team

The person who scopes the work stays accountable for it from the first call to handover, and the team stays on it for the duration, whether that is a short audit or several months. Nobody gets swapped out halfway.

Scope defined before the engagement starts

Broken down into milestones such as modeling, then dashboard building, each with its own out-of-scope line, so the work never blurs into endless iteration.

Price certainty

Fixed price or time and materials, agreed before the work starts and not renegotiated once we are in it.

A weekly feedback cadence

A sync every week, notes in your channel, so problems surface while they are still small.

A project, or an engineer for a few months

Two shapes, both fine. A scoped project, or one named analytics engineer working alongside the project lead at an agreed capacity for three or four months, written into the contract.

Next step

A scoping session

An hour with me. We talk through where your priorities sit right now, what is blocking them, where you could use help, and what support from us would look like. Nothing to prepare, nothing to sign.

Gemma in numbers
20 people, all of whom write code
70+ projects, almost all on dbt
10+ active clients on Lightdash, self-hosted and Cloud

Let's scope the first step together.

Find a time with me