Deblo.ai
AI tutoring for 250 million African students
Educational platform for CP to Terminale with a Pro tier offering 101 specialized AI advisors across 14 business functions (finance, IT, HR, marketing, legal, ops, and more). French-first. Phone + WhatsApp OTP authentication.
Building Deblo
AI Tutoring for 250 Million African Students
The vision behind Deblo.ai: an AI tutor that speaks French, understands the African curriculum from CP to Terminale, costs pennies per question, and works on any phone.
Credits, FCFA, and 6 African Payment Gateways
USD-cent pricing, 13 currencies, Orange Money and M-Pesa, 3 payment gateways, webhook-driven confirmation, and a background poller. Monetizing an African SaaS.
WhatsApp OTP and the African Authentication Problem
Email does not work. SMS is expensive. WhatsApp is universal. How we built authentication for Africa with WhatsApp OTP, Google OAuth, and student access codes.
System Prompts That Teach: Anti-Cheating, Socratic Method, and Grade-Level Adaptation
Six-layer prompt assembly: root identity, school hours detection, grade overlays from CP to Terminale, subject expertise, and exam mode. How we taught the AI to teach.
The Agentic Loop: 24 AI Tools in a Single Chat
Up to 10 iterations of LLM calls, 24 tools from file generation to code execution, background jobs for 30-minute tasks. The agentic heart of Deblo.ai.
Building a React Native K12 App in 7 Days
20 sessions in 7 days: Expo SDK 54 monorepo, Zustand stores, SSE streaming, voice calls via LiveKit, biometric auth, and a custom tab bar.
Generating PDFs, Spreadsheets, and Slide Decks From a Chat Message
Six file generation tools: XLSX, PDF, PPTX, DOCX, HTML, MD. The AI plans the structure, generates via Python libraries, uploads to S3, and streams progress.
AI Memory and Context Compression
Fire-and-forget conversation summarization, the save_memory tool, and context compression at 150K tokens. How the AI remembers across sessions for $0.00005 per summary.
RAG Pipeline: Document Search With pgvector and Semantic Chunking
PostgreSQL with pgvector for embeddings, Datalab API for semantic chunking, Mistral Reranker for top-k refinement, and context compression at 150K tokens.
Background Jobs: When AI Takes 30 Minutes to Think
Queue-bridged architecture with detached asyncio tasks, Redis progress tracking, cooperative cancellation, and stale job cleanup for long-running AI generation.
100 Sessions Later: The Architecture of an AI Education Platform
FastAPI + SvelteKit + PostgreSQL + Redis + React Native. 24+ tables, 100+ endpoints, 60+ components, SSE streaming. The full architecture of Deblo.ai.
Observability: Tracking Every LLM Call in Production
AILog tracks every LLM call. SystemSetting enables config without redeploy. ExerciseResult measures learning. The admin dashboard behind the obscured route.