Back to projects
Professional

Toople AI
Affaq Ahmed · Full Stack / AI Engineer · DevCache · AI / December 31, 2025
- Next.js
- Supabase
- Deepgram
- ElevenLabs
- RAG
- LLM
Overview
Toople AI is an AI-powered sales-training platform designed to improve sales skills through interactive AI coaching. Reps practise and get feedback through natural, voice-based conversations with an AI coach.
My Role & Contributions
I built the full platform from scratch, with a focus on the AI layer:
- Full platform development — frontend and backend on Next.js + Supabase (database and Edge Functions).
- RAG implementation — retrieval-augmented generation to ground the coaching in relevant material.
- Prompt engineering — designed the prompts driving the AI coaching.
- AI coaching workflows — the end-to-end interaction flows.
- Voice — integrated Deepgram (speech-to-text) and ElevenLabs (text-to-speech) for real-time AI voice interactions.
Key Features
- Interactive AI coaching for sales practice.
- Voice conversations — speak to and hear the AI coach.
- RAG-grounded feedback drawn from relevant content.
Tech Stack & Architecture
- App: Next.js with Supabase (database + Edge Functions).
- Voice: Deepgram for transcription, ElevenLabs for speech.
- AI: LLM integrations with a RAG pipeline and custom prompt engineering.
Challenges & Solutions
- Real-time AI voice — combining Deepgram and ElevenLabs into a smooth spoken back-and-forth with the AI coach.
- Grounded coaching — a RAG pipeline plus careful prompt engineering so the feedback is relevant and useful rather than generic.
Timeline
October 2025 – December 2025 · 3 months (MVP)
Links
No public link is available for this project.