Softway Careers | AI Applied Engineer

About the Role

AI thinking becomes working systems.

We have an opening in our team — and we're using it to raise the bar. We're looking for someone who is already thinking and working in AI — not reaching for it when a brief asks for it. Someone who can take a client problem that's unclear, under-defined, or half-formed and architect a system that earns its place in a real workflow. From the first conversation to final deployment.

This is a full-pipeline role. You'll own the problem, the architecture, the build, and the outcome. You'll work directly with clients — which means strong technical execution isn't enough. You'll need the judgment to shape what gets built and the communication to bring stakeholders with you.

This is an in-person role. We believe the best work, the best decisions, and the best teams happen in the same room. Face-to-face, heart-to-heart — that's how we actually operate.

What You'll Do

Real systems. Real clients. Real stakes.

Build agentic systems — multi-step agents that plan, call tools, and recover from failures.
Design and ship RAG and retrieval pipelines over real client data — embeddings, vector stores, hybrid search, reranking, evaluation.
Integrate AI into existing products and workflows — not as a feature layer, but as something that earns its place in how people actually work.
Build evaluation harnesses. Shipping AI without an eval loop is shipping a guess.
Make architecture decisions: model choice, context strategy, memory, orchestration, cost and latency trade-offs.
Own features end-to-end — not just the screen. You'll touch APIs, backend logic, cloud services, and data layers as the problem demands.
Ship in small, fast loops. Question the brief. Care how the thing feels in someone's hand.
Collaborate across Softway's multidisciplinary team — strategy, culture, creative — so what you build is grounded in real human outcomes.
Stay ahead of what's emerging in the AI-native stack and bring it back. We expect you to be reading, building, and sharing constantly.

What We're Looking For

Show us what you've shipped — not just what you know.

We're not hiring for a checklist. But the right person will read this and recognise themselves:

2–5 years of engineering experience with at least one shipped AI system in production — work, side project, or open source. We don't care where. Only that it's real.
Comfortable across the AI stack: LLM APIs, prompting, embeddings, vector DBs, agent frameworks — or the judgment to skip them.
Fluent in Python or TypeScript, passable in the other. Solid on APIs, auth, and enough frontend to stand up a usable interface.
Cloud comfort — deploying to AWS, GCP, Azure, Vercel, Modal or similar without hand-holding.
Engineering discipline — you test what matters, read code carefully, and don't ship what you can't explain.
Able to hold a client conversation — lead a technical discussion, facilitate a workshop, present to a senior stakeholder without losing them.
Treat AI as a collaborator and know when not to use it. That judgment matters as much as the technical ability.

We'll especially notice if you've...

Built a real eval set and used it to make a decision — not just run it as a formality.
Worked with multimodal inputs: vision, audio, document parsing.
Shipped or contributed to something public — open source, technical writing, talks, or visible AI work.
Integrated LLM APIs into something that wasn't a demo.

The Challenge

A real problem. Build something only AI makes possible.

This is the challenge every AI Applied Engineer candidate receives. There's no single right answer. We're evaluating how you think, what you build, and whether you can hold the human in mind alongside the technical.

The Brief

Build a product that could not have existed three years ago. Modern AI is the reason it works — not a feature on top.

Something that is only shippable because of what AI can now do. In your reflection, argue why this product is impossible to build without AI — and why a smarter rule-based or simpler version would fail. Bonus: build the non-AI baseline too, and show the gap.

What This Is NOT

A wrapper. A clever system prompt isn't a product.

An existing app with AI sprinkled on top. If the product worked before AI, it would still work without it.

A faster version of something humans already do. The task isn't new — only the speed is.

What This IS

A product that solves a problem nobody was solving — because nobody could.

Take modern AI away and it's not slower. It's impossible.

What we're asking you to do.

What we're asking you to do

Build a working product

Ship something real. Hosted URL, or a GitHub repo with a README we can run in under 10 minutes. The experience, the logic, the interface — that's on you. It doesn't need to be polished. It needs to work and show genuine thinking.

Write a reflection that shows your thinking

Cover your approach to the problem, how you used AI throughout — from framing to build — the trade-offs you made, and what you'd do differently with more time. This is as important as the code.

Record a short walkthrough video

Max 5 minutes. Walk us through what you built and why. Not a demo script — tell us what you were thinking. What worked. What didn't. What you'd change.

How we evaluate

Problem framing and product judgment

Can you identify a real problem, understand who it's for, and make the case for why AI is genuinely needed — not just added?

Applied AI systems thinking

Can you design a practical AI system with clear reasoning about architecture, retrieval, context handling, evaluation, and trade-offs?

Shipping and engineering execution

Is the build working, usable, and clean? Does it show engineering discipline — structure, deployment readiness, something someone can actually run?

AI-assisted execution

How deeply did AI run through your process — from shaping the problem to building the system? Did you direct it well and own what came out?

Communication and clarity of thinking

Can you explain your decisions, trade-offs, and technical concepts to someone who isn't an engineer?

What to Submit

→ A working build — hosted URL, or a GitHub repo with a README we can run in under 10 minutes

→ A walkthrough video — max 5 minutes (Loom, YouTube unlisted, or similar)

→ Your resume — we evaluate everything together

On AI use: This role is for engineers who are already deep in AI tooling. We're not asking whether you used AI to build — we're looking at how. How early it entered your thinking, how deliberately you directed it, how well you understood and owned what came out. The quality of your AI use is part of what we're evaluating.

📅

Submission Deadline

28th May 2025