AccessLine

Inspiration

AccessLine came from a simple problem: accessibility information online is often incomplete, outdated, or too general to answer a specific question someone has before visiting a place.

I wanted to see if CALL-E could help with that by making a short, transparent phone call to a venue, asking a few focused accessibility questions, and turning the answers into something structured and easy to review.

What it does

Right now AccessLine asks three questions:

• Is there a step-free public entrance?
• Is there an accessible restroom?
• Is there anything about access or arrival that the visitor should know?

The important part is that AccessLine does not try to fill in missing details. If something was not confirmed, it stays unknown.

How I built it

I built AccessLine in Python using the CALL-E Developer REST API and JSON Schema.

AccessLine creates the call, asks the questions one at a time, waits for the result, and then validates the structured response before presenting it.

The CALL-E integration uses outbound call creation, asynchronous call-status retrieval, and structured result extraction.

Challenges I ran into

One controlled real CALL-E test completed successfully and returned a schema-valid result with accessibility information and uncertainty preserved.

I also learned pretty quickly that a call being marked complete by a provider is not the same thing as having a verified usable result.

Two later UX attempts were interrupted by provider-reported signal interference, so I changed the logic to fail honestly when the evidence is incomplete instead of treating an interrupted call as a successful verified result.

What I learned

The biggest lesson was that uncertainty should survive the whole process.

If a venue does not confirm something, AccessLine should not guess.

If a measurement was never provided, it should stay unknown.

Structured data is only useful if it does not make weak evidence look stronger than it really is.

About the demo

The demo video uses a fictional venue and synthetic data so the workflow can be shown clearly without pretending another live call is happening.

The project has separate controlled real-call evidence, but I am not claiming general telephony reliability or broad venue accuracy from one successful test.

What's next

The current version is intentionally narrow. It asks three fixed questions, returns structured accessibility facts, and keeps unknowns visible.

With more authorized testing, the same pattern could potentially be used for other current venue checks where online information is missing or stale.

Upstream CALL-E contribution:

https://github.com/CALLE-AI/awesome-phone-call-agents/pull/284

Built With

• call-e
• json-schema
• python