DataCraft Studios client
AAVision: AI dispatch for an interpreter agency
Email requests become structured jobs, interpreters get offers by link, and AI voice calls handle confirmation and escalation, replacing dispatch by phone and spreadsheet.
- Role
- Architect & lead engineer
- Timeframe
- In production since Sept 2026
- Live
- aavision.net ↗
Problem
Dispatchers handled job requests from email by hand, matching interpreters by phone and spreadsheet.
Approach
Email ingestion via Gmail + n8n → LLM extraction into structured jobs → assignment offers via tokenized links → auto-calendaring, cancellation and re-dispatch → AI voice calls (Vapi) as escalation and confirmation.
Outcome
Saves the agency ~3–4 hours/day already; first operational analytics they’ve ever had.
Architecture
- SourceGmail: Job requests arrive as plain email.Leads to n8n
- Processingn8n: Watches the inbox and hands each request on.Leads to Extraction (Claude)
- ProcessingJobs DB: Single source of truth for every job and its state.Leads to Offer links · Calendar · Vapi calls · Dashboard
- AIExtraction (Claude): Turns free-text email into a structured job.Leads to Jobs DB
- AIVapi calls: AI voice calls to confirm or escalate.
- OutputOffer links: Tokenized accept/decline links sent to interpreters.Leads to Interpreters
- OutputInterpreters: Accept with one click; no login needed.
- OutputCalendar: Confirmed jobs are calendared automatically.
- OutputDashboard: Live operations view and analytics.
Metrics
- Manual dispatch work removed for the agency
- 3–4 h/day
- In production
- Sept 2026
Stack
- Next.js
- Payload CMS
- PostgreSQL
- Neon
- TypeScript
- Anthropic SDK
- n8n
- Vapi
- Resend
- Payload Jobs Queue
- Render