23/06/2026
🔥 Build AI that drives impact
Every company on earth now has the same models you do. Few are getting real work out of them. The few that succeed spend a year doing it, by which point the model they built on has been replaced, and most of the rest ship something that does a fraction of what it should.
The bottleneck was never the model. It's everything above it: the orchestration, the governed knowledge, the wiring into systems people actually use. Most teams rebuild that whole layer from scratch for every project. We built it once, properly. It's called Aerogram, and it means your time goes to the customer's real problem instead of the plumbing.
Your job is to take it into a customer's hardest problem and get it running in production in days, not quarters. A Deployment Strategist owns the customer relationship and brings you the problem; you own how it gets built. The customers are regulated institutions, and their data is the opposite of clean: systems never meant to talk to each other, documents never meant to be read by a machine. The interesting part isn't getting a model to answer once. It's getting a system to answer correctly every time, in a place you don't fully control.
🔥 What you'll do
◾️ Prove it. Build a working proof-of-concept on Aerogram against the customer's real problem and real data, fast. The platform handles orchestration, knowledge, and integration, so you spend your time making the thing work, not rebuilding the foundation.
◾️ Build what the platform doesn't cover yet. Often the highest-value move is a custom interface or integration on top of the stack for one customer, end to end across frontend, APIs, and data. You decide, case by case, whether a problem deserves a durable platform feature or a sharp one-off — and you're right often enough that it compounds.
◾️ Make it real AI, not a demo. This is the hard part. Multi-step agentic workflows that hold up outside the happy path. Retrieval over fifteen years of scanned PDFs no parser handles cleanly. Eval harnesses that prove a workflow is good enough to trust when there's no clean ground truth. No playbook exists for most of it yet, so you'll be writing it. The first version that runs is rarely the one that ships — getting it actually right is the craft.
◾️ Debug what only shows up in the wild. The integration that behaves differently against their data, the edge case their process throws that nobody anticipated. You own it through to working.
◾️ Feed it back. What you discover in the field informs what becomes a durable part of Aerogram and what stays a bespoke build.
🔥 What we're looking for
◾️ A real, working understanding of how LLMs behave. You scope what the model owns versus deterministic code, design retrieval and agent boundaries to fit the data, and when something breaks you diagnose it at the right layer. That judgment matters more than familiarity with any one model.
◾️ 5+ years shipping production software: you stood behind once real people depended on it.
◾️ Full-stack depth: TypeScript with React or Next.js, a backend in Go or Node.js, real database fundamentals (SQL, schema design, migrations).
◾️The ability to make any unfamiliar system work. You drop into an API, codebase, or environment you've never seen, read the docs, and ship against it.
◾️A bias for shipping value over polishing an architecture no one has seen, without dropping quality where it counts.
◾️Thai and English, both with customers.
Strong pluses: Docker and Kubernetes; daily use of LLM coding tools; agentic systems, RAG, workflow automation, or document processing at scale; prior forward-deployed, solutions-engineering, or consulting work.
Don't apply if you need every requirement nailed down before you start, or you'd rather perfect an architecture in private than put a rough version in front of a customer.
🔥 About Cleverse
We're a venture builder: we see where the digital world is going, build the business model around that vision, and craft the product that makes it real. Our teams have shipped the kind of work the industry remembers — built when most thought it couldn't be done at that scale. We host the rooms where the country's builders gather, and we tend to be standing where the field is about to go, not following it.
Aerogram is our current bet — on the gap between AI that demos and AI that does real work inside a business. We stay true to the outcome a customer is paying for, not to any one way of reaching it, and we're never quite satisfied: there's always a version of the work that's better than the one we shipped.
🔥 How to apply
Tell us about something you built that people relied on in production, and what was genuinely hard about it. We read every word.
https://go.cleverse.com/fde-2026
