Nova

Nova : Save money on your electricity bill in 5 minutes

Track: B2C · Team 6 · Paris Fintech Hackathon 2026

Wedge: Electricity · Onboarding: 5 minutes · Avg. savings: ~€150 / yr

Inspiration

Has anyone here chosen to switch their energy supplier this year?

Every year you don't switch, your energy supplier wins. They keep raising prices and it's costing French households up to €500 a year. Yet only 11% of French consumers switched supplier in 2025 (CRE, 2025).

Why? Because most people wrongly believe it's complicated, costly, or risky.

In reality:

No fees
No engagement
No power cuts

French households overpay, and nobody acts. That's why we built Nova.

What it does

Save up to 20% on your electricity bill in 5 minutes with our app.

No phone calls. No paperwork. Just switch.

Nova is an autonomous AI agent that reads your bill, monitors 70+ French offers daily, picks the best deal for your usage profile, and switches you in one tap.

How we built it an AI workflow, end-to-end

Four AI agents. One seamless flow. Each agent has one job, a hard contract, and full audit logs every decision explainable.

#	Agent	What it does	AI inside
01	Onboarding Agent	Reads your bill & your meter a complete onboarding in 5 minute, on the clock.	★ Claude Vision OCR · ★ Profile reasoning
02	Watcher Agent	Monitors the market 24/7 re-scores 70+ French offers every day.	★ CrewAI · Cron · Cloud Functions
03	Decision Agent	Picks the right offer for you. Adapts to your usage and your data and explains every choice the optimizer makes.	★ Tool calling · ★ ML optimization
04	Executor Agent	Switches in one tap. Signs the mandate, triggers the switch, tracks status.	★ Agent autonomy · Audit logs

From bill scan to new contract : fully automated. You do nothing.

Market & business model

The only energy app that makes money by saving yours.


31M	Households in France : our addressable market
€0	The first switch is free
20%	Success fees : we only make money when you do, on savings generated

If Nova doesn't save you money, we don't make a penny. Our incentives are perfectly aligned with yours.

Challenges we ran into

OCR robustness : Real French electricity bills come in 20+ vendor formats: blurry photos, missing fields, rotated PDFs. Solved by Claude Vision in a single pass with strict JSON schema validation.
Explainability vs. ranking quality : A pure ML ranker is opaque, a pure LLM ranker is non-deterministic. We built a hybrid: ML optimization on top of deterministic scoring, with Claude composing the plain-French rationale on demand.
The 5-minute onboarding budget : Bill scan, Linky consent, profile inference, and first ranked offer all had to fit in 5 minutes end-to-end.

Accomplishments we're proud of

A working PWA built in 24 hours.
End-to-end autonomous flow: bill scan → Linky consent → ranked offer + plain-French rationale → one-tap switch.
A multi-agent architecture that scales beyond electricity to every recurring bill.

What we learned

Frontier models change the unit economics of consumer fintech : tasks that needed analysts last year now cost cents.
CrewAI + Claude is a real production stack : agents with tools, memory, and audit logs work today.