Product descriptions:
FinThrive serves thousands of healthcare providers, supporting millions of patients accounts each month across a labyrinth of eligibility checks, claims, and revenue-cycle processes. But the engineering reality was complex: hundreds of Databricks Notebooks, legacy data pipelines, and tribal knowledge scattered across wikis and internal documentation. New engineers spent hours tracing dependencies; experienced team members constantly fielded ad hoc questions from stakeholders and operations. By building a governed, data-grounded knowledge assistant directly on Databricks Agent Bricks, FinThrive made their codebase “askable”. So, engineers and non-technical teams alike can query pipelines in natural language, get cited answers with code excerpts, and deliver innovation with confidence.
Complexity and Context Hidden in a Sprawling Codebase
FinThrive’s data engineering team was actively migrating legacy systems to Databricks while supporting mission-critical revenue-cycle operations for a rapidly scaling customer base. The bottleneck wasn’t the compute; it was context.
FinThrive’s engineering team set out to build clear connections between their source systems, ensuring they could quickly understand what every ingestion job did and how any changes might impact downstream pipelines. They wanted new hires to be able to onboard quickly — confidently identifying relevant tables, formats, and inputs so they could contribute code without delay. At the same time, the team aimed to empower operations and product leadership with greater data visibility, moving the organization toward true self-service and reducing ongoing dependency on engineers for routine questions.
“The manual process was what really slowed us down,” said Ben Bartholic, Principal Data Engineer. “You’d have to know which notebooks to look at, comb through line by line, and try to piece everything together. It wasn’t just a hassle for new team members — everyone lost hours to these detective tasks.”
Building a Governed Knowledge Assistant Where Data Lives
To break through this bottleneck, FinThrive turned to Databricks Agent Bricks, building their knowledge assistant directly on their existing platform and data. Notebooks, repositories, and documentation were exposed via Unity Catalog, allowing engineers to chat with an AI “teammate” grounded in FinThrive’s actual code and data, not static wikis.
“Within a few hours, we had a chat assistant on our pipelines. It felt like turning a static codebase into a living teammate,” said Suhas Byrapuneni, Data Engineer. FinThrive deployed multiple agents — one for Databricks code, another for Azure Data Factory, and a third for internal documentation — then orchestrated their answers through a single supervisor. Any employee could access answers through a Databricks App chat interface.
Engineers ask natural language questions, and the agent retrieves cited code snippets, explains dependencies, and even points out downstream impacts. Prompt flexibility enables both concise and detailed explanations for various situations — ranging from quick checks to complex troubleshooting and in-depth analysis.
What sets this solution apart is its trustworthiness. “We trust Databricks to manage the models behind the scenes, so our team can focus on delivering value to our clients and customers,” said Ben. With Databricks handling updates automatically, engineers avoid chasing model or evaluation upgrades and instead stay focused on innovation. The result: engineers report almost no hallucinations, and answers are consistently accurate as automated Databricks Jobs sync the knowledge source with pipeline updates. Prompt tuning, ongoing system prompts, and robust grounding ensure that responses are both current and context aware.
Faster Engineering, Smoother Onboarding, Fewer Interrupts
While formal measurement is underway, early outcomes are already clear. Tasks that took 15–20 minutes, like listing required job inputs, now yield instant answers — complete with references and code excerpts. Debugging cycles are up to 80% shorter as engineers easily identify “What downstream notebooks use this table?” and jump to the correct code. Onboarding for new team members accelerates, letting them quickly discover tables, formats, and logic without relying on subject-matter experts’ time.
Crucially, non-technical stakeholders — including operations, leadership, and product owners — can now self-serve answers to pipeline and logic questions, freeing engineering from constant interruptions. “Operations leads and product owners can answer common questions on their own, reducing distractions and making everyone more productive,” Suhas shared.
This pattern is extensible: FinThrive will replicate the approach for additional teams as part of upcoming hackathons, expanding business impact across the organization.
FinThrive’s mission is to remove friction from healthcare revenue management. By turning their codebase into a conversational, governed knowledge layer on Databricks, the team is reclaiming engineering time, accelerating migrations, and empowering more people to self-serve answers. As multi-agent orchestration expands and adoption grows, FinThrive expects measurable gains in productivity and time to value, supporting not only internal teams but also the broader healthcare ecosystem.