Founding Engineer for Semantic Search & Conversational AI (Part-Time, Equity-Only) For Global Mental Health

Really Global is building the world's first truly global mental health marketplace, so anyone, anywhere, can find a provider who fits, in their language, at a price they can afford. We launched our prototype in 200+ countries. The foundation exists. Now we need the part that makes it usable at that scale: search that understands what a person means, not just what they typed.

The Opportunity: You Are the Search Team

This is pure sweat equity for an engineer who has already built semantic search and conversational search in production, and built it alone. Not someone who wants to learn search. Not one person on a 5 or 10 person search team. You have designed the whole thing yourself, from data model to retrieval to evaluation to production, and you can do it again.

Here, you own two systems end to end:

1. Semantic search across thousands of providers, many languages, specialties, price points, and formats.

2. A turn-by-turn conversational search assistant that asks a client questions, helps them put what they are looking for into words, turns their answers into a better query and a better set of attributes, and returns search results. It also handles customer-service questions about how the platform works.

We have a working draft of semantic search. You can keep it, rework it, or replace it. That call is yours.

Typical commitment: up to ~10 hours per week, equity-only, alongside your primary role.

Must-Haves (all three)

1. You have shipped semantic search to production at real scale: embeddings, hybrid retrieval, filtering, relevance tuning, and measuring whether results are actually good.

2. You have shipped a multi-turn conversational system to production: LLM orchestration, conversation state, guardrails, latency and cost control.

3. You built at least one of these as the sole engineer, architecture through production.

The Hard Constraint (and why it is interesting)

We are a neutral marketplace. The assistant can ask questions. It cannot give advice, cannot steer, cannot recommend or choose a provider, and cannot triage. Its whole job is to help a client describe what they want so search can do its job. Results are search results, never ranked by who pays. Building a genuinely helpful conversation inside those lines is a hard design problem. If that sounds like a fun constraint rather than an annoying one, keep reading.

Who You Are (and how you show up)

• You build because you can't not. A good problem gets under your skin, and you'll find the hours for it even in a slammed week, not out of obligation, but because shipping the thing is where you come alive. Your agents carry the grind so your energy goes where it counts.

• You are the whole team, and you like it that way. Architecture, data, retrieval, prompts, evals, monitoring, the 2 a.m. fix. You don't need a platform team underneath you or a PM in front of you.

• You measure, you don't guess. You know that "the results look good" is not an eval. You build the test sets and the metrics before you trust a change.

• You've already gone all-in on AI tooling. You run your own setup, your own agents, your own workflows. You're on the premium plans, running multiple agents, because the tooling pays for itself in leverage.

• Maybe you lead now, and you miss this. Your day job moved you into reviews and roadmaps. Part of you misses owning a system end to end with your name on it. This is that thing. (To be clear: here, you're a builder, not a manager. That's the point.)

• Self-propelled. Nobody is going to manage you, and you'd hate it if they tried. The product goals are set. How to get there is on you.

• Candid and low-drama. You can explain a search trade-off to a non-technical founder without condescension, and push back on a bad idea without turning it into a fight.

What You'll Build (and why it matters)

• This is real production software serving real people in 200+ countries, so everything ships carefully, reversibly, and under review.

• Semantic search, end to end. Data and attribute modeling, embeddings, indexing, retrieval, filtering, multilingual handling, relevance evaluation, and production operations.

• Turn-by-turn conversational search. The assistant that asks the right next question, turns answers into attributes, and hands back search results in real time.

• Customer-service chat. The same assistant answering questions about how the platform works, with a clean handoff to a human.

• The guardrails and evals that keep it honest. Test sets, conversation test suites, monitoring, and incident response, so the neutral line holds every time, not just in the demo.

• Our stack: Next.js/React on the front; Python services on the back; PostgreSQL underneath; hosted on Azure; GitHub, Jira, and AI-native workflows (Claude Code, Codex, and whatever you already run). The search and conversation architecture is yours to design.

Your Leverage Is Your Stake

Here's the part most equity deals won't tell you straight: what you earn here is capped only by what you can ship. We measure contribution in Story Points, a unit of complexity and outcome, not hours. Equity comes from a dedicated 5% contributor pool, and it accrues to the people who put work into production. So the engineer who orchestrates agents to ship three times the output earns three times the ownership. Same hours. Different leverage.

• The Deal: Transparent Equity Partnership

• Points for outcomes. Contribution is measured in Story Points: complexity and impact, not hours. Efficiency is rewarded, not punished.

• Merit-based ownership. Points translate directly to equity from a dedicated 5% contributor pool. Your stake grows with everything you ship.

• All real work counts. Search, conversation design, evals, migrations, performance, hardening, documentation, incident response. If it's production value, it earns.

• Earned on shipped value. Work earns when it passes our automated review and is merged. Nobody approves their own work.

• Earn from day one. No cliff. No 90-day pilot. No probation. Your first accepted work earns equity right away.

• You keep what you earn. Points are vested as you earn them. Once you pass 25 Points total, your stake is protected and converts at our first priced round, whether or not you're still with us.

• Real-time transparency. A live dashboard you can check anytime, refreshed weekly, so you always know exactly where your equity stands.

• Locked-in stake. When we close our first priced round, everyone's earned percentage becomes a fixed part of the cap table.

• No hours theater. No politics. Value in production = ownership. It's all in a clear, plain-language agreement covering equity, IP, and security. About 20 minutes of reading.

The Honest Part

This is a side project. You should already have a full-time job, a life, obligations. We respect all of it, and we're not asking you to trade any of it.

What makes this work is leverage and consistency, not raw hours. When there's work, put in real, focused time: up to ~10 hours in a heavy week, sometimes less. That's fine. The one thing that doesn't work here is going dark. So the bar is simple: reply to Slack DMs within 24 hours, keep your Jira tasks current each week, keep your agents working.|

This is not a learning project. If you haven't built production semantic search and conversational search before, this role isn't the right fit, and that's okay.

If you've built this before, alone, and owning the search experience for a platform that could change how a billion people reach care sounds like the thing you've been looking for, APPLY NOW.