HESFB Loan Eligibility Navigator

• 

  hesfb-dashboard

• 

  output

• 

  Output

Inspiration

Thousands of eligible Ugandan university students never apply for the HESFB government student loan, not because they don't need it, but because the eligibility rules are confusing, fragmented, and hard to self-assess. Students miss out on support they deserve simply because the system is too difficult to navigate alone.

What it does

The HESFB Loan Eligibility Navigator is an AI-powered chatbot that reads a student's situation in plain language, reasons through the actual HESFB eligibility rules, and returns a clear verdict "you may qualify, here's why, and here's what to do next." The student types naturally, and walks away knowing exactly where they stand and which documents to gather.

How we built it

Knowledge base: We compiled the official HESFB eligibility rules, approved programmes, approved institutions, required documents, and edge cases (repeating students, dual funding, PWD(People With Disabilities) status, unlisted programmes) into a single structured document the AI reasons over. AI reasoning: Google Gemini reads the student's answers against the knowledge base and produces a structured verdict ; criteria met, criteria unmet, uncertainties, documents needed, and next steps. Responsible AI: Every output uses "you may qualify" framing, never "you qualify." All results cite the source rule and link to hesfb.go.ug. The bot explicitly flags that financial need and final eligibility are determined by the HESFB board, not this tool. Stack: Python, Streamlit, Google Gemini API (free tier).

Challenges we ran into

HESFB eligibility is genuinely complex: first-year vs. final-year rules, STEM vs. affirmative action classifications, regional balance weighting, and PWD provisions all interact in non-obvious ways. Getting the model to reason correctly on edge cases (repeating students, partial bursaries, unlisted programmes) required careful knowledge base design, not just prompting. Responsible AI framing had to be added into every layer; the system prompt, the output formatter, and a post-processing safety net, all to ensure the bot never overstated certainty.

Accomplishments that we're proud of

Built a working AI eligibility navigator in under a week as a two-person student team Scoped to a real Ugandan government system we actually live inside our depth of knowledge showing in the edge-case reasoning The AI correctly handles genuinely complex eligibility logic: first-year vs. final-year rules, STEM classifications, PWD provisions, and dual-funding disqualifications Responsible AI design is added in at every layer, not added as an afterthought

What we learned

Tight scoping wins one real system, one real user, done well, beats a generic tool every time Structuring a knowledge base clearly matters more than prompt engineering; the model reasons well when the rules are clear and unambiguous Responsible AI framing has to be consistent across the system prompt, output layer, and post-processing. Only one layer isn't enough

What's next for HESFB Loan Eligibility Navigator

Expand to cover other Ugandan student support schemes beyond HESFB Add a WhatsApp integration so students can access it on the platform they already use Partner with Makerere University's student services office to deploy it officially Build in multilingual support for students more comfortable in Luganda or Swahili

Built With

• github
• google-gemini-api
• python
• streamlit