Back to projects

AI & Agentic Systems

Study Cost Planner Agent

Plan studying abroad with numbers you can trace to a source.

planning agents
9planning agents
automated tests
93automated tests
API endpoints
25API endpoints
database tables
14database tables

// at a glance

The problem and the idea

The problem

Students compare universities by headline tuition, but the real cost includes living, insurance, visa and transport, and scholarships change the picture completely. The information is scattered, and AI tools often make figures up.

The idea

Nine agents run as a deterministic pipeline over a shared planning context. Costs come from cited sources, every calculation is plain Python, and the LLM only handles language: understanding the request, narrating scenarios and writing summaries.

// who it's for

Who it's for

  • 01

    Students planning a bachelor's or master's degree abroad

  • 02

    Families budgeting for study costs and scholarships

  • 03

    Advisors who need cited, reproducible estimates

// how it works

How it works

intakeSQL+pgvectortuition+livingcurrencyscenariobudget matchscholarshipseligibilitynet valueverifier9 agents · LLM for language, Python for math

1/6 Intake

Swipe sideways to see the whole diagram →

// key features

Key features

  • Total real cost: tuition plus living, insurance, visa, transport and hidden costs
  • Scholarship discovery with an eligible / likely / unknown / ineligible verdict
  • Net-cost ranking after the best award you may qualify for
  • Deadline-aware application plan and a personal tracker
  • Cost radar, cash-flow Sankey and inflation forecast charts
  • Downloadable PDF report with sources cited

// design decisions

Design decisions

  1. 01

    LLM for language, Python for math, so every number is reproducible and auditable.

  2. 02

    The system works end to end without an API key, using deterministic fallbacks.

  3. 03

    Each figure is labelled sourced or estimate, so nothing claims more certainty than it has.

// built with

Built with

  • FastAPI
  • Next.js
  • PostgreSQL
  • pgvector
  • fastembed
  • Docker

Run it locally

$ docker compose up --build