← all challenges

// challenge brief · hack a ton 2026

Bill Buster

Snap a photo of your electricity bill; the agent explains every charge, flags errors, and finds you a cheaper offer.

// Proposed by
Ambasada
// Industry
Energy / consumer
// Difficulty
🟢 Beginner-friendly

// the problem

Romania returned to a free electricity market in July 2025. Prices are set by competition now, not the state, and bills jumped. For the first time, comparing offers and switching genuinely saves money. But bills are opaque (roughly half supply, a quarter distribution, the rest taxes and VAT), so almost nobody does it.

// your mission

Build an agent that reads a photo of a Romanian electricity bill, explains it in plain language, flags errors, compares it against current market offers, and drafts a switch request for the user to approve.

// how it works

  1. 1Read: a vision model extracts supplier, kWh, components, period, total.
  2. 2Explain: an LLM breaks the bill down in plain language and flags anomalies (estimated vs. metered reading, wrong tariff tier, odd jumps).
  3. 3Compare: pull current offers for the user's consumption band and estimate savings, with honest caveats (fixed vs. variable, contract length).
  4. 4Act: draft a switch request. It never auto-submits, and it informs rather than gives financial advice.

// suggested approach

  • Inference: for model access, we suggest [LLMok](https://llmok.app). Use code AMBASADA26 for 50% off.
  • Vision model reads the bill, an LLM explains it, then a small tariff dataset you assemble plus comparison logic does the rest.
  • Focus on electricity (gas stayed regulated longer).
  • Floor version: read one supplier's layout, explain the charges, compare against three offers. Add error detection and multi-supplier robustness next.

// how you'll be judged

  • Extraction accuracy from messy photos.
  • Clarity and correctness of the explanation.
  • Validity of the offer comparison and savings, with no hallucinated tariffs.
  • Robustness across supplier layouts.
  • Creativity, such as consumption forecasting, error detection, and multilingual support.

// stretch goals

  • Forecast next month's bill from usage.
  • Detect billing errors worth disputing.
  • Multilingual mode for non-Romanian residents.

// deliverables

  • Git repo: code + README + run instructions.
  • Demo: a real bill decoded live, with a savings figure.
  • 5-minute pitch: architecture, decisions, trade-offs.