Description
Summary WHO WE ARE We are an agentic-AI consultancy working with mid-market companies in Mexico. We design agents that run alongside our clients' teams, build them, and leave them operating in production. Our house rule: whatever is repeatable runs on deterministic code with hard validations; the LLM only enters where it adds value — the AI proposes, the code executes, a person approves what matters. THE PROJECT An AI recruiting agent for a company that hires operational staff at high volume. It runs on the official WhatsApp Business API (Meta Cloud API) and handles candidates 24/7: answers questions, pre-screens against the criteria of each vacancy, schedules interviews on Google Calendar, keeps a candidate file, and escalates to a human recruiter with operator takeover when needed. It integrates with the client's in-house platform (own API), an ATS (via API or exports), and Google Workspace. Recruiters get a live metrics dashboard. End users are Spanish-speaking candidates in Mexico. YOUR ROLE IN THE 8 WEEKS We run the discovery ourselves. You start with a signed Blueprint in hand: vacancy criteria, conversation guides, sample real chats, integrations defined and access granted. From there you build (weeks 1–4), run a two-week shadow phase where the agent operates in parallel and recruiters supervise it (weeks 5–6), and drive the progressive cutover, live dashboard and closing session (weeks 7–8). During shadow you read real conversations in Mexican Spanish, tune prompts and tone, and iterate directly with the client's recruiting leads — that is why fluent Spanish is a requirement, not a plus. Optional monthly maintenance of the delivered agent afterwards. HOW WE ENGAGE Hourly contract with a weekly hours cap, structured by milestones aligned to the phases above. You work directly with the founder — architecture decisions together, full technical autonomy in execution — and present to the client's team when needed. WHAT YOU WILL BUILD • Conversational agent on the official WhatsApp Business API: business verification in Meta Business Manager, templates, 24-hour window handling, message queues, state machine per candidate, idempotency, human handoff. • Pre-screening logic driven by vacancy criteria (deterministic where possible; LLM for conversation, triage and edge cases). • Google Calendar scheduling and Google Workspace integration (Drive, mail) with OAuth2. • Integrations with the client's in-house platform API and ATS; semi-assisted job posting to Facebook groups. • Candidate records and a live metrics dashboard (response time, cycle time, no-show). • Deployment in the client's own cloud accounts (AWS/GCP or VPS), with logging, monitoring, backups and kill switches. Personal-data handling compliant with Mexican privacy law (LFPDPPP). DEFINITION OF DONE • Agent live on a client-owned, Meta-verified WhatsApp number, handling real candidate conversations end to end in Spanish. • Pre-screening decisions traceable to vacancy criteria; every escalation to a human logged with reason. • Interviews scheduled on Google Calendar without double-booking; candidate file updated automatically. • Integration with in-house platform and ATS running on the client's infrastructure with monitoring and alerts. • Runbook, architecture doc and a training session for the client's recruiting team delivered. • Two-week shadow period passed with recruiter sign-off before cutover. REQUIREMENTS • Fluent or native Spanish (you will read real candidate chats, tune the agent's tone and work directly with a Mexican recruiting team) and working English for our team. • Production Python: robust services (queues, state machines, idempotency, validation, logging) — not notebooks. • Real integrations: OAuth2, webhooks, SQL, Google Workspace APIs; WhatsApp Business API in production (Meta Cloud API or a BSP), ideally having done the Business Manager verification yourself. • LLM APIs (Claude, OpenAI or Gemini) with tool use, bounded prompts and cost control per outcome; evals are a plus. • Deploy and operate on the client's infrastructure: Linux, Docker, cron/systemd, VPS or AWS/GCP in client accounts. • Security and privacy discipline: compartmented credentials, auditable and reversible actions. • Full self-management and clear communication with a client team. NICE TO HAVE Playwright/browser automation against systems without API · TypeScript/MCP · RAG · voice agents · prior consulting work facing clients. HOW TO APPLY Answer the screening questions with specifics. We read every proposal; generic AI-written cover letters are the fastest way to be skipped. Please include one link (repo, demo or write-up) that shows a messaging agent or backend service you built and operated.