Iustis

Inspiration

We're law and engeneering students, and long before this hackathon we kept seeing the same scene play out: someone with almost no legal knowledge, standing in front of a police officer or a boss, having to decide in real time whether to sign something, pay someone off, or push back. The numbers back up what we'd already noticed. In Mexico, 93.4% of the crimes committed in 2025 were never reported or investigated (INEGI, ENVIPE 2026), and separately, 82% of Mexicans say access to justice feels unjust to them (Latinobarómetro, 2021). That gap isn't abstract to us: it's the distance between what the law says and what people actually get to use, and it's usually widest in the first five minutes of a problem, before anyone has time to call a lawyer.

[Espacio dejado a propósito: si tú o tu socio tienen una anécdota o motivación personal concreta detrás de esto, va aquí. No la inventé porque no la sé.]

What it does

Iustis is a first-contact legal triage assistant that runs over WhatsApp (or SMS), so nothing needs to be installed and it works on a weak connection. A user describes their situation by text or voice note (a traffic stop, a dismissal letter). The assistant classifies the area and urgency, asks for whatever information is missing, and answers in a fixed, plain-language format: what is happening, your rights (with the article quoted), what to do right now, where to go, and when the situation actually needs a lawyer.

Two design rules hold the whole thing together. First, every legal claim is grounded in a retrieved article from the official text; if nothing relevant is retrieved, Iustis says so and refers the user onward instead of guessing. Second, numbers are never left to the model: a deterministic calculator computes labor amounts like severance and shows the formula step by step.

The prototype goes deep on two modules, traffic law in Mexico City and labor dismissals, built on an architecture where adding a third area of law (constitutional, civil, criminal) means loading a new corpus, not rewriting the pipeline.

How we built it

[WhatsApp sandbox / web chat simulating it] → webhook ([FastAPI]) → [speech-to-text for voice notes] → a triage classifier that outputs structured data (area, urgency, missing information) → retrieval over a legal corpus split by article and organized as independent modules per area of law → [the LLM] drafts the answer strictly from the retrieved text → a deterministic calculator for labor amounts → [a PDF generator] for a short summary or draft document.

Tools and libraries used: [Python, FastAPI, el LLM/modelo usado, el método de vector store o búsqueda, el servicio de speech-to-text, el proveedor de sandbox de WhatsApp, hosting]. Declarar aquí también cualquier herramienta de IA usada para programar o redactar, tal como pide el criterio de "Presentation & Documentation" del hackathon.

Challenges we ran into

Keeping the assistant inside what the retrieved text actually supports, instead of sounding confident when it shouldn't. Scoping two legal areas deep (traffic in Mexico City, labor dismissal) rather than five shallow ones, inside a seven-day window. Drawing a defensible line between legal information and legal advice, which we handle with an explicit notice and a referral path instead of silence. And building a WhatsApp-first channel without depending on Meta's Business API approval landing in time, which is why the prototype falls back to [a sandbox / a web chat that mirrors WhatsApp].

Accomplishments that we're proud of

A retrieval rule with no exceptions: no article retrieved means no legal answer, not just in principle but enforced in the pipeline. A labor calculator that shows its math instead of asking the model to compute it. A modular corpus structure that lets a third area of law be added without touching the retrieval or triage code. And a problem statement backed by verified data from both Mexico and the wider region, not just assumption.

What we learned

[Una lección real y específica de construir o probar el prototipo, algo que los haya sorprendido o que harían distinto ahora. No la puedo escribir por ustedes.]

What's next for Iustis

More areas of law and more jurisdictions inside the same modular architecture. A pilot with a legal clinic. The official WhatsApp Business API for a real deployment beyond the hackathon prototype. And longer term, the access-to-justice gap this addresses isn't unique to Mexico, so the same architecture, modular corpus, citation-grounded answers, a channel that needs no install, is built to extend across Latin America.

Built With

• html
• phyton