Ask Moses: Technical Architecture

AI-powered sales coaching platform built on modern tech stack. Real-time analysis, instant feedback, continuous improvement.

Tech Stack

Modern, scalable, and production-ready

⚡

Next.js 16

Full-stack React framework with App Router

🔷

TypeScript

Type-safe development

🎨

Tailwind CSS v4

Utility-first styling

🟢

Supabase

PostgreSQL + Auth + Storage

📦

Vercel Blob

Large file uploads up to 50MB

🧠

AI SDK

Vercel AI SDK for LLM integration

📧

Resend

Email delivery service

Database Schema

PostgreSQL via Supabase with RLS policies

rubrics
Coaching configurations & system prompts
idnamesystem_promptllm_modelis_active
scripts
Sales process templates with sections
idnamesections (JSONB)criteria (JSONB)rubric_id
calls
Recorded analysis results
idtranscriptcriteriascoreemail_sentscript_id
criteria
Evaluation framework
idnamedescriptionsort_orderrubric_id

API Routes

Server Actions & Route Handlers for backend logic

POST /api/upload-audio
Large file upload handler (Vercel Blob)
Client requests upload token
Validates file type/size (max 50MB)
Returns signed URL for direct upload
Bypasses 4.5MB serverless limit
POST /api/transcribe
Audio to transcript conversion
Receives Blob URL
Fetches audio from Blob storage
Uses OpenAI Whisper
Returns cleaned transcript
POST /api/analyze
AI call analysis engine
Fetches script + criteria
Gets system prompt
Calls GPT-4o-mini or Gemini 2.5
Returns structured feedback
POST /api/generate-criteria
Auto-generates criteria from script
Analyzes script description
GPT creates criteria items
Saves to database
POST /api/send-coaching
Sends branded email feedback
Takes analysis results
Generates HTML email
Sends via Resend

AI Integration (Vercel AI Gateway)

Multi-model support via unified provider API

Available Models

openai/gpt-4o-mini

Fast, cheap, reliable

google/gemini-2.5-flash

Balanced speed & quality

google/gemini-2.5-pro

Most powerful analysis

How It Works
1. Script Context

AI receives sales process template

2. Analysis

Evaluates transcript vs script + criteria

3. Feedback

Structured JSON with scores & tips

Data Flow

How a call gets analyzed end-to-end

1

Upload Audio/Transcript

Trainer submits call via dashboard

2

Store in Vercel Blob (if audio)

Client uploads directly to Blob (up to 50MB)

3

Transcribe via Whisper

API fetches from Blob, sends to OpenAI Whisper

4

Fetch Script + Criteria

Get scoring rubric from Supabase

5

AI Analysis

GPT/Gemini evaluates against script

6

Save Results

Store in Supabase calls table

7

Send Email

Branded HTML feedback via Resend

Evolution Roadmap

From MVP to revenue-driving features

Phase 1.5: Twilio
85% → 95% automation
  • Real-time call recording via Twilio webhooks
  • Auto-transcription on call end
  • Immediate email trigger
  • No manual upload needed
Phase 2: GHL Integration
Multi-tenant ready
  • Per-team scripts via GoHighLevel
  • Sync results back to CRM
  • Workflow triggers for follow-ups
  • Multi-source ingestion (Twilio + GHL + manual)
Phase 3: Advanced Analytics
Revenue-driving features
  • Team comparison dashboards
  • Trainer progress tracking
  • A/B testing scripts
  • Predictive coaching insights

Ready to Deploy?

The MVP is production-ready. Start with Phase 1.5 (Twilio) to eliminate manual uploads and automate the entire flow.

Go to DashboardView Presentation