Clarity

• 

  Clarity

Inspiration

Job applications often contain claims that sound specific but still leave an important question unanswered.

A résumé might say someone “maintained 98% CSAT,” but was that their own score or the team’s? A static form cannot know which follow-up question to ask.

We built Clarity around a simple idea: instead of collecting more application text, use a short adaptive phone call to clarify the one thing that actually matters.

What it does

Clarity reads the job description, résumé, and written answers, then identifies a job-relevant claim that is worth clarifying.

It calls the candidate, identifies itself as an AI assistant, asks the initial question, and can adapt with one follow-up based on the candidate’s answer.

In our demo, the application says “98% CSAT.” During the call, the candidate explains that 98% was actually the team metric. Clarity then asks for their personal score and learns it was 96% across about 45 surveys.

The result shows the clarified fact alongside the candidate’s own words and transcript evidence.

Clarity does not score candidates, rank them, or make hiring decisions.

How we built it

Clarity is built with Next.js, React, TypeScript, Gemini, and CALL-E.

Gemini analyzes the application and proposes job-relevant ambiguities. Before a question can be used, Clarity verifies that the claim is anchored in the original application and filters questions involving sensitive personal attributes.

CALL-E receives the opening question, context for the conversation, instructions for an adaptive follow-up, and a structured result schema.

After the call, Clarity normalizes the CALL-E response and turns the transcript and structured output into a simple before-and-after clarification.

We also built replay and debug modes so the entire product flow can be tested without placing real calls or spending API credits.

Challenges we ran into

The hardest part was deciding what Clarity should not do.

It would have been easy to turn it into another AI candidate scoring system. Instead, we kept the scope focused on clarification and left judgment to the human hiring team.

We also had to make adaptive phone conversations predictable enough for a product workflow. That meant carefully defining the opening question, limiting the follow-up, requesting structured results, and preserving transcript evidence.

Another challenge was safety. Questions need to stay tied to job-relevant claims and avoid sensitive personal attributes, so we added claim anchoring and question filtering before a call can happen.

Finally, we needed a reliable way to develop and demo a phone-based product. We built synthetic replay fixtures, debug flows, call inspection tools, and a shared normalization pipeline so replay and live calls behave as similarly as possible.

Accomplishments that we're proud of

We are proud that the final experience became extremely simple:

one ambiguous claim → one short phone call → one clearer fact

The demo shows that the follow-up question does not have to be known in advance. Clarity learns that “98% CSAT” was a team number, then adapts and asks for the candidate’s personal score.

We are also proud that the evidence remains visible. Clarity can show what the candidate originally wrote, what was asked, what they answered, and what remains unknown.

Most importantly, Clarity helps humans get better information without pretending that AI should make the hiring decision.

What we learned

We learned that the most useful AI workflow is not always the one that automates the biggest decision.

Sometimes the better role for an agent is to notice uncertainty, ask the right follow-up question, and return better evidence to a human.

We also learned that phone calls become much more useful when they are adaptive. The value is not simply turning a questionnaire into speech. The value is being able to ask a different second question because of what someone just said.

And from a product-design perspective, we learned that showing one clarified fact is much more powerful than presenting a large AI-generated report.

What's next for Clarity

Next, we want to make Clarity useful beyond the single demo scenario.

That means improving ambiguity selection across different job types, making call flows more configurable, strengthening safety controls, and supporting production-ready authentication and storage.

We also want to explore the same pattern outside hiring: situations where written information is incomplete, but one short adaptive conversation could turn an ambiguous claim into a useful fact.

Built With

• adaptive-calling
• call-e
• clarity
• evidence-backed
• gemini
• human-in-the-loop
• next.js
• react
• structured-results
• typescript