Ally Co
Voice-first roleplay training for real-life allyship (Gemini + ElevenLabs + Next.js)
Live demo: https://allyco.vercel.app/
The problem
A lot of people genuinely want to be supportive when they see bias or harm happening, but in the moment they:
- freeze
- don’t know what to say
- worry they’ll escalate things
- or only think of the “perfect line” three hours later in the shower
Allyship is a skill. Skills get better with practice. So we built a place to practice.
What Ally Co does ?
Ally Co gives you realistic scenarios and lets you practice your response using voice roleplay, then get coaching-style feedback.
On the home page, users can choose a scenario like:
- Gender Bias (someone gets subtly undermined in a meeting)
- Racial Microaggressions (a “joke” crosses a line in a group setting)
- Misgendering (wrong pronouns, then everyone moves on)
Each scenario has a mission goal, like:
- “Back them up without escalating the room.”
- “Address it professionally, keep dignity centered.”
- “Correct politely and support the impacted person.”
There’s also an Open Voice Chat mode where you can talk through a real situation and get guidance.
And because dopamine is a core framework now, we added Badges & Streaks:
- “Gender Bias Champion”
- “Dignity Defender”
- “Respect Advocate”
Key features
- Voice-first practice
- Scenario-based missions with clear goals
- Real-time coaching feedback (focused on practical, repeatable behavior)
- Open Voice Chat for personal situations
- Badges + completion tracking to encourage follow-through
- Daily Streaks to encourage people to keep coming back everyday.
Tech stack
This project is built with:
- Next.js (frontend + app structure)
- Gemini (LLM for roleplay + feedback)
- ElevenLabs (voice)
- Vercel (deployment)
The goal was a smooth loop:
- User speaks
- Model responds in-character
- App gives coaching feedback
- Repeat until it feels natural
Product philosophy: “Practice > perfection”
We tried to avoid making this feel like a lecture. The point is not “say the single correct sentence.” The point is:
- speak up sooner
- stay respectful
- center the impacted person
- keep the room from exploding
So the feedback is designed to be coaching-style, not judgmental.
Where Ally Coach can be used ?
Ally Coach is built for any place where people want to respond better in real time but don’t get to practice.
Schools and Universities
- Orientation or first-year seminars: practice speaking up in group discussions and residence life situations
- DEI workshops: replace passive slides with active roleplay
- Student leaders, RAs, clubs: rehearse interventions that stay calm and respectful
Workplaces and HR training
- Manager training and onboarding: practice addressing bias without escalating the room
- Team culture and psychological safety programs: build “say something sooner” habits
- Remote teams: voice-first practice for awkward Slack and meeting moments
Nonprofits and community programs
- Volunteer training: practice responding to harm in public-facing situations
- Youth programs: confidence-building for respectful communication under pressure
The goal is simple: practice responses before the moment happens, so people freeze less and act sooner.
What was hard (because it always is) ?
- Latency + real-time feel: voice apps feel broken if timing is off by even a little.
- Keeping scenarios realistic: too dramatic feels fake; too mild feels pointless.
- Feedback that’s actionable: “be better” is useless. The app needs to tell you what to try next.
What’s next ??
If we keep building Ally Co., the next upgrades are:
- more scenarios (workplace power dynamics, public settings, friend groups)
- difficulty levels (gentle vs direct responses)
- better progress tracking (strengths over time, not just completion)
- a “replay” mode to practice alternative responses
Try it
If you’ve ever thought, “I should’ve said something,” this is for you.
Live demo: https://allyco.vercel.app/
Log in or sign up for Devpost to join the conversation.