Sara

Problem

On a fractional real estate platform, sales and customer success have the same conversation all day. How does fractional investing work. What would this unit return. Which property fits my budget. That front half is repetitive, and it eats the hours the team should spend closing. It is also where all the useful context about a user surfaces, then evaporates.

Role

Built solo on ElevenLabs and Claude. Sara answers investing questions, calculates ROIs from the user's inputs, and suggests units to fit their style, flip or hold. It drives the UI: redirecting to units, highlighting property cards, opening the investment modal. Each transcript yields six data points, budget to objections, so the team gets structured context for manual follow-up.

Outcome

Sara is a prototype. The shaping decision was what to automate and what to leave human: it takes the repetitive front half and leaves the close to the sales team. Six evals score every conversation on ElevenLabs. Two exist for the failure modes: brevity, because rambling kills a voice agent, and guardrail compliance, no promised returns, no tax or legal advice.

Stack