Hataf Atif

Orenda Project · Taleemabad

A FastAPI + Supabase service that exposes lesson-plan generation as an API, with @dars/client and @dars/react so regional teams integrate against a key instead of forking a codebase. The chapter planner is buffer-budgeted and re-plannable: holidays and exam periods change the plan rather than invalidating it. Lesson plans generate per slot, on demand.

• ›Chapter Planning Engine — LLM planner grounded in explicit planning principles
• ›Buffer-budgeted, re-plannable class paths; holidays and exam periods as first-class inputs
• ›Per-slot lesson plans and formative assessments generated on demand
• ›Multi-tenant org API keys — including the cross-tenant key-leak fix on signup
• ›Playwright e2e harness, Bruno request collections, ADRs for the decisions that stuck

Orenda Project × National Institute of Education

WhatsApp Cloud API bot: interactive list menus, Flows for multi-select quiz questions, and session state that survives a force-close in the Android wrapper. Training rules — levels, attempt ceilings, pass marks, cooldowns — resolve from one source of truth instead of three. Certificates are PDFs minted on demand; R2 media is presigned inline so training videos play in the thread.

• ›Training levels with exams, MSQ quizzes over WhatsApp Flows, and varied questions per attempt
• ›Certificate PDFs minted on demand and delivered in-thread
• ›Reflective coaching conversations, with slash commands that reliably end them
• ›Portal sessions that survive a hard refresh and an Android force-close
• ›Play releases, versionCode discipline, named release artifacts

Orenda Project · Taleemabad

A skill pack that gives every team's coding agent working Slack and Notion access — per-team tokens, ten-minute setup. Rewrote slack-search to route point lookups directly rather than iterating: roughly 20% fewer tool hops at equal correctness. Also a worldmap skill that de-collides labels and picks label contrast per region automatically.

• ›Per-team token model — no shared credentials, no central bottleneck
• ›slack-search point-lookup routing: ~20% fewer hops at equal correctness
• ›worldmap skill: automatic label de-collision and per-region contrast

An MCP server that decodes mixfont ghost-font videos. The message is hidden in a field of random dots — every individual frame is pure noise, and the text only exists in the difference between frames. It recovers the signal across time, then serves the answer as a tool an agent can call.

• ›Temporal integration across frames to recover a signal absent from every still
• ›Ships as an MCP server — your agent calls it, you don't run anything by hand
• ›Self-installing: point an agent at SETUP.md and it registers itself

A Slack + Notion + GitHub harness that runs my personal projects. It provisions its own Notion database schema on setup, then reconciles issues, agent runs and status back to a single board.

• ›Provisions the full Notion database schema during setup
• ›Slack as the control surface, GitHub as the source of truth

EFT

A DynamoDB → RDS ETL pipeline, a serverless function extracting parts-of-speech and named entities from customer messages, an analysis layer on top of the result, and a websocket endpoint feeding realtime WhatsApp into a CSR inbox.

• ›easydost_etl — DynamoDB to RDS, scheduled and idempotent
• ›eft_pos_ner — serverless POS + named-entity extraction over message text
• ›EasyAnalyst — the analysis layer the extraction fed
• ›whatsapp-chat-backend — serverless websocket endpoint for the CSR inbox