First Line

Inspiration

Crisis line volunteers are everyday people stepping up to answer calls from people at the worst moment of their lives. Since the 988 lifeline launched, contact volume has jumped over 80%, leaving crisis centers struggling with volunteer recruitment and retention. Up to 20% of crisis workers consider quitting due to stress, anxiety, and feeling unprepared.

The first time a crisis line volunteer hears "I don't want to be here anymore" shouldn't be from someone who actually means it.

Right now, training relies on static manuals and awkward peer roleplay. Peer roleplay falls short because another volunteer can't convincingly sound like a person in active distress, simulate a breaking voice, erratic breathing, or sudden hostility on command. Volunteers finish training feeling unprepared, and when their first real shift hits, the fear and emotional weight trigger immediate burnout. Volunteers don't quit because they lack compassion; they quit because we throw them into life-or-death situations without realistic practice.

What it does

First Line gives volunteers a safe, hyper-realistic sandbox to panic, mess up, and build muscle memory before a real life is on the line.

• Realistic Audio over Text: Powered by ElevenLabs, the caller's voice actually breaks, hesitates, or gets defensive. Trainees learn to respond to tone and emotional cues, not just plain words on a screen.
• Persistent Context & Replays: Powered by TigerData, session state stays intact. If a volunteer mismanages a high-stakes call, they can re-run the exact scenario with their previous history remembered to try a better approach.
• Guarded Persona Scenarios: Backboard strictly controls our caller personas so simulations stay clinically grounded without breaking character or hallucinating.
• Clear Readiness Analytics: Our dashboard replaces vague "good job" feedback with exact data. Trainees see their thresholds, risk assessment scores, and missed warning signs like hopelessness or perceived burden.

How we built it

We built First Line with a light, fast tech stack focused on low-latency voice interaction and session memory:

• Backboard: Handled scenario logic, dynamic caller personas, and state transitions.
• ElevenLabs API: Streamed expressive, emotionally variable audio directly to the trainee.
• TigerData: Stored context, memory, and caller history across multiple training runs.

Challenges we ran into

• Long response lags ruin the feel of a real phone call. We spent hours cutting down latency while keeping the vocal emotion intact.
• Crisis scenarios cannot have wild hallucinations. Backboard strictly guardrails the caller behavior to keep every scenario safe and realistic.
• Turning unstructured conversations into reliable risk scores was tough. We reworked our evaluation prompts until the feedback felt accurate and constructive.

Accomplishments that we're proud of

• We skipped flat text chats to build a voice caller that sounds scared, hesitant, or overwhelmed in your headset.
• Callers remember past interactions. If you make a mistake, you can retry the scenario and see how your new approach changes the outcome.
• Instead of generic advice like "be empathetic," our app flags specific missed cues and gives a clear path to fix them.

What we learned

• How someone speaks completely changes the right way to respond.
• Volunteers need to see their exact triggers and missed cues rather than a simple pass or fail grade.

What's next for First Line

• Partner directly with 988 training directors to align our metrics with official evaluation standards.
• Add calls covering substance abuse, teenage distress, and severe trauma.
• Build a dashboard where trainers can track volunteer readiness and step in when extra support is needed.

Built With

• backboard
• elevenlabs
• express.js
• figma
• gemini
• node.js
• react
• render
• tigerdata
• typescript