ClaimFlow - A multi call orchestrator for evidence gathering specific to auto insurance domain

Inspiration

The process of claiming an insurance takes a lot of calls to parties involved, insurance providers, third parties such as police, towing companies, repair shops, hospitals etc involved to fulfill one claim. I saw an opportunity to automate few calls that could save both the party and provider a lot of time. Most of the calls made for evidence gathering is a repeated process and can be automated. More over, no insurance related skills or apps found in the repository and hence I want to add a valuable, meaningful work flow to this repo.

What it does

This library adds value to this repo by making a call to the claimant to gather as much as evidence required to proceed in the claim process. I narrowed down scope to claiming insurance in the auto insurance domain. It makes two calls to the claimant. One call is made to gather detailed evidence about the incident and asks for the policy number. The second call is made only when the backend determines that the policy related to the policy number is valid and covers the incident. The second call is to notify that an adjuster is assigned to the claimant and adjuster will be in touch in stipulated amount of time. The second call is not place if the policy number is invalid and/or the policy doesn't cover the incident, meaning no adjuster is assigned.

How I built it

I used chatgpt and claude to brainstorm the idea and narrow down the scope to generate specifications. I used KIRO IDE to specify the requirements to generate the code base.

Tech stack:

• Node.js + TypeScript
• CALL-E Phone API (call orchestration)
• Jest (unit testing)

Challenges & Lessons Learned

The challenge is to define the use case and its purpose so that this project add meaningful and valuable substance to this repository. Insurance domain is an ocean and has millions of use cases. I had to narrow down to fit only one use case for this hackathon. So the challenge was choosing the reliable use case and at the same time a managing this project to be not too complex. With the tools that I had, I was successful to define a short scoped reliable and valuable use case.

Key Takeaways

• Multi-call workflows are the next frontier in phone automation. Single-call interactions are limited; stateful multi-call patterns unlock complex processes.
• Constraints breed creativity. JSON Schema limitations forced us to think harder about nullable field representation. The empty-string sentinel actually simplified testing.
• Safety policies are features, not bugs. The 422 error redirecting us away from policy collection was a hard constraint-but it also made our solution safer and more trusted.
• Conversation design matters as much as data design. Question order, pacing, and context affect both user experience and data quality.
• Dry-run first, live later. Starting with a no-call mode let us iterate fast, test thoroughly, and deploy with confidence.

Built With

• call-e
• jest
• node.js
• typescript