This isn't a job. It's a mission.
One billion people are struggling with their mental health. 700,000 preventable deaths each year. Only a fraction ever reach the right provider because borders, languages, and broken marketplaces stand in the way.
Unacceptable. We're fixing it.
Really Global is building the world's first truly global, provider-centric mental health marketplace — so anyone, anywhere, can find the right human in their language, at a price they can afford. We're already live at really.global in 200+ countries. The foundation exists. Now we layer in the intelligence that makes finding the right provider feel like talking to a wise friend — not filling out a form.
The Opportunity: Own the AI Layer of a Global Mental Health Marketplace
This is pure sweat equity for a senior applied AI engineer who wants meaningful ownership in a global platform — alongside their primary role. Typical commitment: 5–10 hours/week, equity-only. The model layer, the retrieval stack, the conversational architecture — they're yours to shape.
You'll own the architecture and the builds — from retrieval systems to real-time conversational matching to the agentic tools our providers will use to care for their clients at scale.
Most AI product roles hand you a narrow feature inside someone else's platform. This one hands you the entire AI surface of a company.
We're backed by founders with a track record of scaling ventures to $130M+ in revenue. This will be the most meaningful AI work of your career.
Who You Are (and how you show up)
- You've shipped AI in production, at a real company, to real users.
Not demos. Not notebooks. Not side experiments. You know what it's like to watch a RAG system degrade, a chatbot hallucinate at the wrong moment, or an agent loop in an expensive failure mode — and you know how to prevent, detect, and recover from all three.
- You have production experience in at least two of these areas:
retrieval systems / semantic search / RAG; real-time conversational AI (voice or text) with LLM orchestration; agentic systems with tool use and structured outputs. You're hungry to grow into the third.
- You take AI in healthcare seriously.
You understand that a chatbot helping someone in crisis find a provider is not the same as a chatbot helping someone return a sweater. Hallucination, tone, safety, and escalation are live concerns — not afterthoughts.
- You build, you don't wait.
You see the roadmap and you start shipping. You don't wait for specs or ask for permission.
- Pragmatic, not perfect.
You know when to fine-tune, when to prompt, when to use a bigger model, and when to ship the dumb version that works. You default to evaluation-driven development because vibes don't scale.
- Clear, candid communicator.
You can explain model trade-offs to non-technical founders, and give direct feedback to engineers shipping AI features that aren't production-ready.
- AI-native workflow.
You orchestrate your own work with tools like Claude or Codex — not because it's trendy, but because anything else is slower.
What You'll Build (and why it matters)
This is the roadmap. All of it needs to work in production, for real people, in a healthcare context where hallucinations and bad recommendations have real consequences.
- Semantic search and discovery
— help someone describe what they're struggling with in natural language and surface the right providers across languages, specialties, price points, and modalities. RAG done properly: embeddings, reranking, evaluation, and the hard work of measuring whether results are actually better than keyword search.
- Conversational matching and support
— real-time conversational AI that handles customer support at scale and acts as a guided matching experience, talking a user through their needs and surfacing the right provider in the right language, in the moment. Latency, cost, safety, and escalation-to-human are all first-order problems here.
- Provider-side agents
— personalized client engagement, homework assignments, structured surveys, intake summaries, progress dashboards, session prep. Give every provider on the platform leverage they've never had — without replacing the human relationship that makes therapy work.
- Real-time audio translation
, longitudinal client insights, outcome measurement, and whatever comes next. The AI surface of a global mental health marketplace is vast and largely unwritten. You'll help shape it.
Our stack:
Next.js/React (frontend); Python services (backend); PostgreSQL; hosted on Azure; Cloudflare