Biu

The Problem. In Brazil, over 11 million people are unemployed, and millions more are locked out of the formal job market not because they lack skills, but because they lack the digital literacy to present them. Building a CV requires typing into forms, writing in professional register, and navigating job portals — all insurmountable barriers for workers with low education who have spent years cooking, cleaning, selling goods informally, or doing manual labor. Their experience is real, but the tools to translate it into a professional document were never designed for them.

Biu directly addresses UN Sustainable Development Goals 8 (Decent Work and Economic Growth) and 10 (Reduced Inequalities) by removing the structural barriers that prevent informal workers from accessing formal employment — turning voice into economic opportunity.

Our Solution. Biu is a WhatsApp voice assistant that turns a natural conversation into a professional, print-ready CV and real job leads — no typing, no forms, no apps to install. The user sends an audio message saying “oi” and Biu responds with a warm, friendly voice, guiding them through six conversational stages: personal details, education, work experience, skills, extracurriculars, and career objective. When the user says something like “fazia bolo e vendia pros vizinhos” (“I used to make cakes and sell them to neighbors”), Biu’s AI transforms this into Harvard-format professional language: “Produziu e comercializou aproximadamente 30 bolos por semana, gerenciando pedidos e atendimento ao cliente.” Once confirmed, Biu generates a formatted PDF curriculum and immediately searches Brazilian job portals (Indeed, Catho, Gupy, SINE, Vagas.com) for matching openings, delivering clickable links right in the chat. The entire experience happens inside WhatsApp — the app 99% of Brazilians already use daily.

Creativity and Innovation. Biu’s core innovation is treating voice as the primary interface for professional document generation. Instead of forcing low-literacy users to adapt to text-heavy tools, we bring the tool to them on the platform they already know. The AI doesn’t just transcribe — it performs a semantic upgrade from colloquial, informal speech to structured professional language using action verbs and quantified results, a task that previously required a human career counselor. The combination of multimodal speech understanding, language transformation, PDF rendering, and live job search grounded in real web results — all orchestrated through a single WhatsApp conversation — is, to our knowledge, novel.

Social Impact and Transformation. Biu directly addresses employment inequality by removing the barriers between informal workers and formal job opportunities. A domestic worker, a street vendor, or a construction laborer can now have a Harvard-format CV in their hands within minutes, along with real job links tailored to their profile and city. The user can delete all their data at any time by saying “apagar meus dados,” and automatic retention policies ensure nothing lingers beyond 90 days — full LGPD compliance. By meeting people where they are (WhatsApp, voice, Portuguese), Biu makes professional self-presentation accessible to those who need it most but have been systematically excluded from digital employment tools.

Technical Feasibility and Execution. Biu is a fully functional prototype, deployed as two Docker services communicating over an internal network:

Gateway (Node.js 20 + whatsapp-web.js + headless Chromium): manages the WhatsApp session, classifies incoming media, serializes messages per phone via a FIFO queue with 120-second timeouts, and delivers replies (audio, PDF, text).
Brain (Python 3.12 + FastAPI): orchestrates conversation state (SQLite in WAL mode), calls Google Gemini for all AI tasks, and renders the CV PDF.
AI is the backbone of every step:

Gemini 3 Flash (multimodal) receives raw audio bytes directly — no separate speech-to-text service — and returns structured JSON with the user’s transcript, Biu’s response, and conversation metadata in a single API call.
Gemini 3 Flash + Google Search Grounding searches the live web for real job postings on major Brazilian portals, with anti-hallucination safeguards that reject shortened URLs and fall back to pre-formatted portal searches if results are insufficient.
Gemini 2.5 Flash TTS (voice “Algenib”) synthesizes warm, natural Brazilian Portuguese audio responses, converted to OGG Opus for efficient WhatsApp delivery. URLs are stripped before synthesis so the voice never reads out links.
Jinja2 + WeasyPrint renders the Harvard-format CV from a structured Pydantic model into a clean, print-ready PDF.
The system is resilient by design: every downstream failure (PDF rendering, job search, TTS) degrades gracefully to a friendly text reply — the /message endpoint never returns a 500. Structured logging with phone masking, exponential-backoff retries, and a 16-test pytest suite ensure production-grade reliability. Deployment requires only Docker and a Gemini API key.

Team Collaboration. Our team combined domain research (understanding the real barriers low-literacy workers face) with rapid AI-assisted prototyping. We used Claude Code for architecture design and implementation, while Gemini powers all runtime intelligence. The system prompt — the “personality and rules” of Biu — is mounted as a read-only volume, meaning we can iterate on conversation flow, tone, and language transformation rules without rebuilding any container. This allowed us to refine Biu’s behavior through fast feedback loops with real conversational scenarios.

The entire system — from architecture to deployed prototype — was built from scratch during the hackathon. Our commit history reflects continuous development across all stages, from the initial WhatsApp gateway to AI orchestration, PDF rendering, and job search integration.

Everything was created during the hackathon, except the start of the project’s scoping.