Your360 AI: LiveKit and State Machines for 360 by Interview

Your360 AI explains how its vocal coach Tam orchestrates a 360-degree feedback process, from collection to debriefing. The team emphasizes strict anonymity rules, multi-pass evaluations, and complete control over the voice stack, with a median debriefing time of 44 minutes.

Anonymity and Quality of Output Governed by Rules and Two Passes

Your360 AI mandates corroboration for critical feedback and only displays groups with three or more providers. The team specifies that the synthesis must remain accurate, corroborated, and truly anonymous. To ensure quality, two evaluation passes are applied: the first, structural, aims to detect errors such as misattributed feedback; the second, substantive, re-engages the analysis with fewer rules to identify patterns that may have been missed in the first pass. The debriefing is conducted through gradual revelation so that the recipient addresses one theme at a time, and the feedback aligns with the values and leadership principles specific to each company.

Management of the Voice Stack: Migration to LiveKit and Model Selection

Your360 AI has transitioned from Retell to LiveKit to control speech-to-text conversion, the LLM, and voice synthesis. To structure the conversation, the team replaced a single prompt with state machines, which also manage transitions between strengths, growth areas, and specific requests. According to Your360 AI, reasoning models are not suitable for voice, and model exchanges are mediated between personality and time until the first draft.

Reasoned Observation of Usage and Automation of Alerts

Your360 AI discusses finding a balance between observability and the protection of sensitive data. The company recommends collaborating with design partners who agree to access conversations at launch, then implementing automated alerts so that human intervention only occurs in the event of detected anomalies. Monitoring focuses on sentiment, customer value, and product quality rather than a systematic review of transcripts. Usability tests combine screen recordings and follow-up interviews to compare perception with what actually happened.

Product Process in Three Calls and Duration Metrics

The journey includes three appointments with coach Tam: a personal discovery, the collection of feedback from respondents, and then the debriefing of the report. The first debriefing call exceeded the 60-minute limit, while the median duration now stands at 44 minutes. Users thus engage for an average of 44 minutes during this median debriefing.

Iterations, Collection, and Evaluation by Personas from Failures

Your360 AI indicates that it launched and validated the personal discovery agent one month before deploying the collection, then added the synthesis. Personas derived from failed calls serve as the basis for evaluations and regression testing during model and provider upgrades. Collection calls are optimized for completion and low friction, including via phone. In voice interactions, individuals can answer three questions right from their first sentence, which requires the agent to track what has already been covered and decide when to delve deeper or move on. The service offers a 360-degree feedback process distinct from annual evaluations and survey-based 360s, and aims to transform awareness into action between the first and second cycles.

Team Visualizations and Organizational Use Cases

Your360 AI provides organizational heat maps, usable for interdependent teams as well as independent cohorts. The collection calls are designed to minimize friction, and these deliverables target various usage contexts within organizations.