Basata scales healthcare voice agent testing 6x faster with Hamming

Basata's agents are configured per practice, each with its own conversation flows, scheduling rules, and guardrails. Writing and maintaining test cases by hand for every configuration was not sustainable.

With Hamming's API, the team defines agent behavior in YAML and converts it into test cases programmatically. That lets them auto-generate test cases from configured agent functionality, manage state across multi-step conversations (for example, checking that an appointment was actually booked), inject scenario-specific test data into each run, and keep tests in sync with configuration changes.

The programmatic layer removes the need to create or update test cases by hand in a UI.

Every practice Basata serves has its own agent with unique conversation flows and conditionals. Testing every path by hand took individual phone calls, one at a time, with one person on each.

With Hamming, the whole suite runs concurrently, so the team starts every test at once instead of placing serial calls. Concurrency matters because each agent has its own conditionals, LLMs are stochastic so tests must run several times for confidence, and the number of tests grows with each new customer.

That frees the team to focus on building rather than testing.

Basata serves a diverse patient population. The team has seen regional differences in voice and communication style change how patients respond to the agent, with some combinations causing immediate hang-ups.

Hamming's persona library covers demographics the team cannot replicate by hand. Switching genders, accents, and speaking speeds doesn't scale with a human team, but the platform has it built in: accents and regional speech patterns, speaking speeds and communication styles, voice and gender combinations that trigger different patient responses, and demographics that match real patient populations.

Next, the team plans to A/B test voice and persona combinations in production, using Hamming's data to find which configurations work best for different patient demographics and regions. That moves them from reactive fixes to proactive optimization based on real data.

Basata's most sophisticated use of Hamming goes beyond testing. The team feeds the descriptions from Hamming's LLM judge evaluators back into a local agent that tunes system prompts, model selection, and agent configuration automatically.

The loop runs the test suite through the Hamming API, collects evaluator descriptions and failure analysis, passes the results to a local agent that updates the system prompt, then re-runs the tests to confirm the improvement.

To the team, manual prompt engineering felt less like engineering and more like trial and error. The loop replaces that guesswork with data and lets the agents improve themselves.

Testing dropped from 90 minutes of manual phone calls to 15 minutes of automated concurrent runs. The team starts every test at once and reviews the results, and the time saved compounds with every new customer deployment.

Before Hamming, every development iteration needed another round of manual calls, and the social and cognitive drain of talking to agents took energy from engineering. Now the team makes a change, runs the suite, reviews results, and repeats, without picking up a phone.

The agentic feedback loop turns Hamming's evaluator descriptions into agent configuration changes automatically, improving agent quality while cutting human effort to near zero.