Summer of Code
Build Something Real.
The entry point into Afrium. We have defined 5 specific, nationally significant projects upfront. Train with us, prove your worth, and build a real AI system for Pakistan alongside the Afrium team.
10 Seats Per Project
50 total seats across 5 projects. Highly focused, small groups. No matter where you are from, we select based on hunger to learn, not credentials.
Beginners Welcome
Never built an AI tool? Apply anyway. The Afrium team will guide and lead you step-by-step through the entire technical process.
Open to Everyone
College students, undergraduates, and graduates are all eligible. You don't need a specific degree or university name to build here.
The 5 Internal Projects
What You'll Build
Claude Skill: HEC Pakistan
Build a reusable Claude Skill that teaches any AI coding agent how to interact with HEC's degree attestation and university recognition systems. Every Pakistani student going abroad needs HEC attestation. Nobody has built AI tooling for it.
Real World Example
The problem: Navigating HEC's complex requirements manually.
What you walk away with
A published open-source .md Claude Skill on GitHub that any developer in Pakistan can use. First Claude Skill ever built for a Pakistani government system.
MCP Server: NADRA
Build a working MCP server that connects NADRA's public-facing services — CNIC verification, family registration certificate, death certificate — to any LLM. Developers and businesses currently have no clean way to integrate NADRA data.
Real World Example
The problem: Integrating identity verification into local apps is painful.
What you walk away with
A live open-source MCP server on GitHub that any Pakistani developer can connect to Claude or any other LLM. First MCP server ever built for a Pakistani government institution.
Agentic CLI: Pakistan Legal Research
Build a terminal agent that takes a plain English legal question — 'what are my tenant rights in Punjab' or 'what does Section 144 actually mean' — breaks it into research steps, pulls from Pakistan's legal database, and returns a clear structured answer.
Real World Example
The problem: Millions face a severe legal literacy crisis.
What you walk away with
A working open-source CLI agent on GitHub. First agentic legal research tool built specifically for Pakistani law.
Multi-Agent System: Pakistan News
Build a multi-agent system where specialised agents monitor Dawn, Geo, ARY, Tribune simultaneously — one agent per source — then a master agent synthesises them, removes duplicates, detects bias differences, and produces a clean daily intelligence brief.
Real World Example
The problem: Media bias and overwhelming noise in daily news.
What you walk away with
A live open-source multi-agent news system on GitHub. First multi-agent AI system built specifically for the Pakistani media landscape.
Context Engine: Urdu Long Documents
Build a context engineering system that handles long Urdu government documents — budget speeches, court judgements, policy papers — managing what gets included, compressed, and retrieved so an LLM can answer questions accurately. Pakistan's government runs on long documents nobody reads.
Real World Example
The problem: LLMs hallucinate or lose context on 200-page Urdu PDFs.
What you walk away with
A working open-source context engineering pipeline on GitHub specifically built and tested on Pakistani Urdu government documents. First of its kind anywhere.
How It Works
Your Selection Process
Application & Screening
Free · 30 min
Tell us who you are. No GPA required, no resume. You'll take a short logic and problem-solving test. We select the top 50 students (10 per project) based purely on logic and hunger to learn.
Community Training
Weeks 1-3
The 10 selected students in each project group train together for 3 weeks. During this time, the Afrium team watches closely — who is active, who asks smart questions, who shows up, and who is serious about building.
The Proposal
Week 4
Submit a technical proposal aligned to your specific project. We usually already know who we want based on the first 3 weeks — the proposal just confirms it. One student per group is finally selected for the build phase.
The Build
Weeks 5-8
The selected student builds the system directly with the Afrium core team for 4 weeks. The final output isn't a toy project — it is a real, nationally significant AI system published for all of Pakistan.
The Output
What a Summer of Code output looks like
Track 02: MCP Server
NADRA Identity Verification MCP
{
"mcpServers": {
"nadra": {
"command": "npx",
"args": ["-y", "@afrium/mcp-nadra"]
}
}
}The Output
A complete Model Context Protocol (MCP) server that connects Claude Desktop directly to the NADRA public identity verification API.
Impact
Built over the 48-hour assessment window. Features automated rate-limiting, secure token storage, and context-aware schema parsing. Published on npm and GitHub. This server allows any local LLM to natively query NADRA data using natural language.
This is not a certification.
You won't get a PDF. You won't sit through lectures. You won't watch recorded videos. You will find a real person, understand their problem, and build something that works. When you're done, you'll have a live URL, a real user, and a story worth telling. That is worth more than any certificate.