2024 - Present
1.Adaptably

Project details:
EdTech / Learning Centre ManagementLearning platform for the Grade Solution learning centre: students work through exercises matched to their level, earn reward points, and redeem them. AI is built into the app for marking, handwriting OCR, oral exam scoring, and a chat on learning progress for students and parents.
Technologies
Full-stack Engineer
About 3,000 users, 800 of them active regularly. AI runs both inside the product and in how it is built and operated.
Achievements
- Integrated AI marking for over 1,000 worksheets a month, with a pre-check that drops empty, junk, and copied answers before any LLM call; teachers review only disputed marks, so results arrive far sooner.
- Gave each AI feature (marking, OCR, oral exam scoring, chat) its own model and harness to keep cost down.
- Developed with a multi-agent setup: Claude as the lead agent orchestrating Codex and Grok as sub-agents, taking each new request through to test and review within the coding standards.
- Tracked errors in Sentry and took user feedback directly through Chatwoot into an automated AI workflow, so issues are fixed and answered quickly and accurately.
- Built an AI analytics chat where students and parents ask about learning progress, streamed over SSE with per-role access checks and a daily rate limit.
- Built homework OCR: scanned PDF pages become images so AI can read handwritten answers and marker ticks.
- Built the whole backend alone for about 3,000 users: the PostgreSQL data model on Directus and Node.js APIs (Directus extensions, Express) for levelled exercises and reward points, with Twilio and CleverTap.


































