All case studies
Early-stage B2B SaaS · data tooling

Built the API middleware that let a B2B SaaS offer data integration as a first-class product feature.

40+
Customer integrations running through the middleware in production
1
Clean API surface between their frontend and Airbyte's complexity
<200ms
Median middleware response time at the API layer
OAuth
Connection flow handled end-to-end for their customers
Day 1
Independent deploy cadence handed to client engineering
Client context

An early-stage B2B SaaS company whose product needed to offer data integration capabilities to their customers. Building that from scratch wasn't their core competency, so they needed someone to stand up the infrastructure that wrapped Airbyte's API into something their frontend could call cleanly.

The problem

Airbyte gave them the integration engine, but their product needed a polished layer between their frontend application and Airbyte's API. Going direct at Airbyte meant exposing too much complexity to end users and too many failure modes to support. They needed OAuth flow handling, source and destination configuration, normalized request and response formats, and an admin interface their team could operate. On top of that, they needed BigQuery destinations with GCS staging as the default loading method, tested and documented for their customers.

The scale of the problem mattered as much as the shape of it. The team was already shipping their core product roadmap; they didn't have the bandwidth to take a six-month detour into integration infrastructure. They needed someone who could land a working middleware quickly, transfer it cleanly, and not leave behind a black-box dependency that would slow them down for the next year.

What we built
01

Foundation: FastAPI app + data layer

A FastAPI application deployed on Heroku with uvicorn serving and SQLAlchemy handling the data layer (users, OAuth connections, source configurations, structured logging). Starlette underneath manages sessions and the admin UI. Authentication, error handling, and request shaping live at this layer so the rest of the system stays clean.

02

Integration: Airbyte API wrapper

Normalized request and response patterns sit between their frontend and the Airbyte backend. The destination router handles BigQuery with full GCS staging configuration (bucket names, paths, HMAC credentials, keep-files behavior) and surfaces sensible error messages when Airbyte's API returns one of its many failure modes.

03

Operations: admin UI + CI/CD

An admin interface gives their team real-time visibility into Airbyte API calls and responses, so support can actually debug customer integration issues. GitHub Actions ship changes to Heroku on every merge to main, with environment-specific configuration so their team owns the deploy cadence end-to-end.

How we worked

The Blueprint phase ran in three weeks: scoping the middleware contract with their frontend lead, mapping the OAuth flow end-to-end across the source platforms they cared about, and identifying the destination configurations needed for launch. We came out with a tight API surface document and a sequenced backlog so the Build could ship value every two weeks rather than landing as one big handoff at the end.

Through the three-month Build, we paired with their backend engineer in twice-weekly working sessions, with code review running asynchronously on GitHub. We deliberately mirrored their existing engineering patterns (Python conventions, repo structure, deploy approach) so the middleware would feel native after handoff rather than like a foreign body their team would resent maintaining.

Knowledge transfer was the deliverable, not an afterthought. By month two, their backend engineer was reviewing our PRs and merging them. By month three, he was opening his own PRs against the codebase. We delivered a runbook covering common failure modes, the admin UI for debugging, and a CI/CD pipeline configured for their team. The day we wrapped, they shipped the next feature themselves.

Results
  • Clean API surface their frontend team built against without touching Airbyte directly
  • 40+ customer integrations running through the middleware in production
  • BigQuery destination with GCS staging available as a first-class option for their customers
  • Admin UI gave their team real-time visibility to debug customer integration issues
  • CI/CD to Heroku so their team could ship changes independently
We needed a polished surface between our product and Airbyte's complexity, and we needed it without sacrificing six months of our roadmap. They built exactly that, and our frontend team has never had to touch Airbyte directly.
Co-founder & CTO, B2B SaaS client

Outcomes start with a Blueprint. We plan, build and run from there.

Thirty minutes with a 829 Analytics partner. You leave with a prioritized view of what to build first, what's worth waiting on, and the business metric anchoring each move. Whether or not we end up working together.