Case Study
Laser
Data Infrastructure
Starburst · Starburst Enterprise Platform (SEP)

AI-Accelerated
Enterprise Platform Upgrade

How Light Craft Labs partnered with Starburst's engineering team to compress a complex SEP version upgrade — using GenAI tooling to reduce migration risk, accelerate documentation, and ship ahead of schedule with the full team upskilled.

3wk
Ahead of Schedule
0
Production Regressions
100%
Team Upskilled
Faster Documentation

A mission-critical upgrade with real complexity

Starburst is the company behind the Starburst Enterprise Platform (SEP) — a leading federated query engine that allows enterprises to run high-performance SQL analytics across multiple data sources without moving data. SEP underpins analytics infrastructure for some of the world's largest organizations.

Starburst's engineering team was facing a major SEP version upgrade — a project involving dependency migrations, API surface changes, connector compatibility validation, and extensive documentation updates across a large, mature codebase. The complexity was real: version drift had accumulated, test coverage was uneven, and the team needed to maintain full production reliability throughout the process.

The organization had the technical talent. What they needed was a force multiplier — a way to compress the upgrade timeline, reduce the cognitive load of migration decisions, and ensure the entire team came out the other side more capable, not just the codebase.

What made this upgrade hard

01

API Surface Changes & Dependency Drift

Significant changes in the SEP API surface between versions meant manual impact analysis across a large codebase — a slow, error-prone process that traditionally requires deep familiarity with every module.

02

Connector Compatibility Risk

SEP connects to dozens of data sources. Each connector required compatibility validation against the new version — a testing surface that would traditionally demand significant manual QA investment.

03

Documentation Gap

A mature, evolving codebase inevitably accumulates documentation debt. The upgrade created a forcing function to close those gaps — but doing so manually would have added weeks to an already tight timeline.

04

Team Knowledge Distribution

Not all engineers had equivalent depth in every affected subsystem. The team needed a way to level up quickly — so the upgrade didn't create new knowledge silos.

How we worked

We embedded with Starburst's engineering team and ran a Laser engagement — hands-on, in the work, using GenAI tooling to compress every phase of the upgrade cycle.

1

AI-Assisted Impact Analysis

We used LLM-powered code analysis to map the full blast radius of API surface changes across the codebase — surfacing breaking changes, deprecation paths, and dependency chains in hours rather than days. This gave the team a prioritized migration backlog from the start, not after weeks of manual discovery.

Codebase scanning
Breaking change detection
Prioritized migration backlog
2

Automated Migration Scaffolding

For high-volume, pattern-based migrations — function signature changes, import path updates, configuration schema adjustments — we built GenAI-assisted tooling that applied changes systematically across the codebase, with human review on edge cases.

Pattern-based automation
Human-in-the-loop review
Regression test generation
3

LLM-Accelerated Documentation

We deployed a structured documentation pipeline — using LLMs to generate inline documentation, update changelog entries, and produce connector compatibility matrices from test outputs. Engineers reviewed and refined, rather than writing from scratch.

Inline doc generation
Compatibility matrix synthesis
Changelog automation
4

Embedded Knowledge Transfer

Alongside the technical work, we ran a parallel upskilling track — pairing junior engineers with AI-assisted exploration tools so they could reason through unfamiliar subsystems independently.

AI-paired onboarding
Subsystem walkthroughs
Workflow playbooks

What shipped — and what changed

Shipped Ahead of Schedule

Completed ~3 weeks ahead of original timeline estimate

Zero Production Regressions

Every change tracked and validated, no incidents

Full Team Upskill

Every engineer left with practical fluency in AI-assisted dev workflows

4× Faster Documentation

LLM-assisted pipeline closed significant tech debt backlog

"

Ashutosh onboarded 3× faster than employee onboarding, and shipped MVP for Starburst Platform Upgrade automation in just 4 weeks. We are looking forward to continue to work with Light Craft on this core component of our strategic customer retention priority.

Jitender Aswani SVP of Engineering, Starburst Data, Inc.

Have a complex technical project to accelerate?

Whether it's a migration, a new build, or a system that needs to move faster — we work in the work with you.

Book a Discovery Call
← View All Work