DeskMate AI

Your all-in-one AI assistant for calls, bookings, and customer engagement.

Comment

Inspiration

DeskMateAI began with a simple moment inside a tiny threading parlor in Queens. The owner—an older Indian woman with limited English—did everything herself: pick up the phone, manage walk-ins, track loyal customers in her head, and still deliver great work.

Her business didn’t struggle because of demand. It struggled because she didn’t have the time or tools meant for businesses like hers.

DeskMateAI was built for her and millions like her. The solo esthetician, the nail tech, the barber, the lash artist—people who keep neighborhoods running but rarely get access to modern technology. We wanted to build something that acts like a silent teammate, powered by AI, that handles the busywork automatically.

What it does

DeskMateAI is an AI assistant for calls, bookings, and customer engagement for small businesses. Under the hood, it runs a constellation of specialized AI agents that coordinate like a real team—handling conversations, scheduling, customer insights, and business intelligence.

1. AI Voice Receptionist (ElevenLabs + OpenRouter Agents)

Every customer call is streamed to an agentic LLM via OpenRouter.

ElevenLabs performs high-fidelity speech → text.

Our “Receptionist Agent” interprets intent, asks clarifying questions, verifies availability, and collects booking details.

A “Validation Agent” checks for errors or missing data.

A “Calendar Agent” finalizes the slot and schedules with Google Calendar.

The entire call feels natural, responsive, and human.

2. AI-Driven Booking + Calendar Integration

Once confirmed, bookings automatically sync to the business’s Google Calendar. The agents handle:

conflicts

time zone logic

duration calculations

confirmation messages

The small business owner doesn’t touch anything.

3. Real-Time Owner Dashboard

A clean dashboard centralizes:

appointments

revenue metrics

customer history

engagement data

low-demand predictions

monthly insights

All powered by analytics models running against structured Supabase data.

4. Engagement Agent (Twilio + OpenRouter)

This AI agent monitors service repeat cycles, overdue visits, low-demand windows, and customer patterns. It automatically drafts:

reminders

follow-ups

personalized messages

targeted promotions

And sends them via Twilio with owner approval.

5. Cross-Reference Customer Intelligence

A dedicated agent analyzes customer overlap across businesses on DeskMateAI. This unlocks:

cross-promotions

referrals

loyalty bundles

partnerships between local businesses

Something small businesses normally can't do without deep analytics.

How we built it

DeskMateAI's architecture is built around modular agents, each responsible for a specific part of the workflow, all orchestrated through OpenRouter for model routing.

Voice Pipeline (Real-time AI)

ElevenLabs for streaming speech → vectorized text.

OpenRouter to route each agent to the most appropriate LLM (4o-mini for intent parsing, Qwen for reasoning, etc.).

Twilio for the telephony layer.

Handling real-time voice in a multi-agent system—and keeping conversations natural—was one of the hardest parts.

Agent Architecture

We implemented separate agent personas:

Receptionist Agent (conversation + extraction)

Validation Agent (semantic checking + error recovery)

Calendar Agent (availability, conflicts, Google Calendar logic)

Engagement Agent (behavioral targeting + message generation)

Analytics Agent (demand prediction + segmentation)

Cross-Reference Agent (customer overlap + partnership suggestions)

Each agent communicates through structured JSON messaging—strict schemas enforced through model instructions.

Backend

Supabase for relational data + row-level security

Time-series analysis for demand forecasting

PostgreSQL triggers for event logging and behavioral analytics

Frontend

React + Vite + Tailwind

Real-time dashboards

Engagement panel for AI-generated campaigns

Calendar visualization

Insights feed generated by the Analytics Agent

Challenges we ran into

Real-time LLM inference over a phone call while preserving conversational flow

Preventing agents from hallucinating or drifting from schema

Merging disconnected customer records across businesses reliably

Ensuring that ElevenLabs → OpenRouter → Twilio pipeline didn’t introduce latency

Designing an agent-to-agent communication protocol that can’t break during live calls

Making the whole thing simple enough that an overworked salon owner can trust it instantly

Accomplishments we’re proud of

A fully autonomous AI receptionist that schedules real bookings from real calls

A cross-agent orchestration system that behaves like an actual operations team

A flexible voice pipeline combining ElevenLabs, OpenRouter, and Twilio

A dashboard that finally gives small businesses insights they’ve never had access to

A system that feels genuinely helpful, not gimmicky

What we learned

Speech interaction requires more empathy than accuracy—timing, tone, and pacing matter.

Agentic systems only work when each agent is extremely focused and tightly constrained.

Small businesses want automation that feels invisible, not overwhelming.

Well-structured data unlocks insights that even the business owners didn’t know existed.

What’s next for DeskMateAI

WhatsApp conversational booking

Multilingual voice agents for immigrant-owned businesses

Predictive staffing suggestions

Demand-based smart pricing

A marketplace for cross-business loyalty bundles

A mobile app for owners on the go

DeskMateAI is evolving into a fully autonomous AI teammate for small businesses—one that works alongside them, grows with them, and removes the operational load they shouldn’t have to carry alone.

Built With

Share this project:

Updates