PowerPlea

Inspiration

Rodolfo, a busy attorney, was spending dozens of hours per petition manually cross-checking every claim, exhibit, precedent and statute—an expensive, error-prone process that begged for automation.

What it does

PowerPlea ingests a petition and its supporting files, extracts each asserted claim and its cited evidence, then automatically:

Verifies whether the evidence truly supports the claims via a RAG-powered review of exhibits
Identifies and evaluates relevant precedent cases
Checks applicable statutory provisions
All findings are overlaid on the petition text with color-coded highlights indicating strong support, weak support or gaps.

How we built it

We built PowerPlea on Toolhouse.ai, leveraging its agent orchestration to streamline our workflow. A FastAPI backend triggers a core “claim extraction” Toolhouse.ai agent, then fans out three parallel Toolhouse.ai agent groups—evidence, precedent, and statutory—each powered by vector-search retrieval on the supplied documents. Outputs from these agents feed into our React frontend for real-time annotation and interactive review.

Challenges we ran into

Containerizing our stack proved tricky due to an opaque, “vibecoded” Dockerfile we hadn’t authored.

Accomplishments we’re proud of

We shipped a fully functional prototype with a polished frontend and end-to-end pipeline—and still managed to sleep! Tackling a genuine legal bottleneck validated both our architecture and the promise of agent-based workflows.

What we learned

Toolhouse.ai dramatically accelerates building multi-agent apps, but “vibe coding” without clear ownership in critical infra can lead to deployment headaches. Clear separation of concerns and early code review are essential.

What’s next for PowerPlea

We’ll broaden input types—starting with an image-analysis agent to verify visual exhibits—plus integrate structured legal databases, benchmark accuracy, and roll out fine-tuned models for jurisdiction-specific precedent and statutory interpretation. We want to see whether we can validate the product and gain traction.