AI-Mer

Inspiration

Dating apps create fatigue: too much swiping, ghosting, and low-signal conversations. We wanted to remove that friction and help people get to meaningful compatibility faster. The core idea was to automate early-stage screening with AI agents that represent each user’s authentic personality, values, and dating goals.

What it does

AI-Mer is an agentic dating platform that replaces random swiping with a structured compatibility pipeline.

Users complete voice-first onboarding to create a rich profile. The system stores profile and conversation context, then initializes a personalized AI agent. Two agents run a real-time speed-dating simulation. A live scoring engine tracks compatibility as messages are exchanged. Users receive a match report with trend-over-time compatibility and a final score. If the match is strong, users can choose to move forward and plan a date.

How we built it

We built AI-Mer as a full-stack, real-time system:

Frontend: React/Next.js for onboarding, live conversation UI, and compatibility visualization. Backend: Node.js/Express with Socket.IO for real-time orchestration and messaging. Data layer: MongoDB Atlas for onboarding transcripts, profiles, vectors, and interaction history. AI layer: Gemini API for goal-conditioned agent behavior and profile/personality extraction. Voice layer: ElevenLabs for high-fidelity onboarding interaction. Scoring: custom real-time ScoringEngine with weighted components and EMA smoothing. Compatibility is computed from four weighted pillars:

Pre-conversation alignment (30%) Personality alignment (25%) Emotional flow (25%) Topic alignment (20%) To reduce score volatility during live chats, we apply exponential moving average smoothing (alpha = 0.3), then display a stable trend graph.

Challenges we ran into

Keeping the scoring loop truly real-time on every message without UI lag. Preserving user authenticity while still allowing autonomous agent behavior. Making compatibility scores stable and interpretable instead of noisy. Integrating voice, AI, persistence, and live sockets into one reliable pipeline. Designing hard-constraint gating so recommendations are actionable, not vague.

Accomplishments that we're proud of

Built an end-to-end agentic dating workflow from onboarding to match decision. Implemented per-message compatibility telemetry and a live trend-over-time graph. Created a system that combines qualitative voice context with quantitative match signals. Replaced swipe-based guesswork with a transparent, data-driven match report. Demonstrated a scalable architecture for real-time, AI-mediated compatibility screening.

What we learned

Voice onboarding captures emotional nuance better than static profile forms. Real-time transparency increases trust more than a single final “black box” score. Compatibility quality depends on both semantics and conversational dynamics. Clear system boundaries (capture, persist, simulate, score) are essential for reliability. Agentic products require strong orchestration and telemetry, not just good prompting.

What's next for AI-Mer

Improve scoring fidelity with deeper emotional and long-term behavioral signals. Add memory-aware agent refinement for stronger user representation over time. Expand explainability so users can see exactly why compatibility increased or dropped. Strengthen scheduling and post-match workflows for smoother transitions to real dates. Continue scaling toward a platform focused on authentic connection at scale.