About the job

TL;DR: Own the machine-learning systems at the core of a live, AI-native fashion product (self-hosted and frontier models, computer vision, and our auto-linking pipeline), and ship plenty of product and backend work alongside them. This is a founding-level role with a path to fully owning AI and a team. We care more about how fast you can learn than about what you already know.

The Role

This role is roughly half AI/ML and half product engineering, and that split is deliberate.

For the AI half, you'll own the models and pipelines that power the core Lekondo experience: self-hosted LLMs, computer-vision segmentation, embeddings and retrieval, and the auto-linking system that turns a photo of an outfit into structured, deduplicated wardrobe items. You'll have real latitude here, including training or fine-tuning custom models where it earns its keep.

For the other half, you'll ship like the rest of the team. We're small and there's a lot to build, so you'll write backend code well outside of ML: optimizing our Go services, attacking performance, and shipping features end to end. You'll likely lean backend, but if you want to reach into the mobile app and go full-stack, that is welcome too.

As we scale, the AI half grows: you'll fully own the AI roadmap and build and lead the team underneath it. For now, you're a hands-on builder who happens to own the most strategically important surface in the company.

What You'll Own
• Model strategy. Lekondo runs a mix of frontier APIs and open-weight models (think Gemma and Qwen) on our own GPUs, across both AWS and GCP. We process thousands of outfit uploads a day, and each one fans out into a burst of segmentation, embedding, classification, and generation calls, well over 100,000 model inferences daily that all have to stay fast and cheap as we grow. You'll own the routing that sends each job to the right model, the fallback behavior that keeps the product up when a model (or a whole cloud) has a bad day, and the cost and quality tradeoffs across the entire mix. In short, you'll decide when to self-host, when to call a frontier API, and when to train something custom.
• Auto-linking (our flagship ML system). The hardest, most-iterated system we have: a hybrid pipeline of multimodal embeddings, vector search, and LLM re-ranking that matches each garment in a new photo against a user's existing closet. You'll own its precision and recall, its decision thresholds, and its evolution.
• Computer vision. Text-prompted segmentation (SAM3), background removal, image generation, and face anonymization, all running on our own GPU clusters across AWS and GCP. You'll own model selection, quality, latency, and cost.
• Model quality and evals. The classifiers that read aesthetics, colors, taxonomy, and composition from an outfit. You'll own the prompts, the few-shot calibration, the evaluation harnesses, and the relentless quality tuning that separates "demo" from "production."
• The other 50%: product and backend. We're small and there's a lot to ship, so you'll write plenty of code that has nothing to do with ML. That means real backend engineering in Go: the API on ECS, the fleet of Lambdas and Step Functions behind our async pipelines, caching and read-path optimization, and the data modeling that keeps everything fast at scale (DynamoDB & OpenSearch experience is a real plus). You'll ship user-facing features and chase performance wherever it matters. And if you want to reach into the React Native app and go full-stack, that is fair game too.

How We Build

We are an AI-native engineering team, and we mean it:
• We don't hand-write code. Claude Code, Codex, or whatever tools get the job done, and engineers have full autonomy over their setup.
• Humans own system design and review. Agents write the functions; you own the architecture, the full context, and the PR review. That is where the real leverage, and the hardest thinking, lives.
• Our scale demands judgment. We're past the point where the simplest one-shot solution scales. We're deeply performance-focused, and we need you to know what we're building toward so we're not refactoring around a clever-but-wrong shortcut every day.
• You learn on the job, with AI. You don't have to arrive an expert. You do have to be the kind of person who can take an ambiguous goal in a domain you've never touched, research it, talk to the codebase, prototype and ship it with AI, and then own it over the following months until you are the expert. Learning live as we deploy and optimize is the job.

If you've felt the shift in how software gets built over the last two years and you're hungry to operate at the very top of it, you'll feel at home here.

How We Work
• High trust, high ownership. We won't set your hours; you own results. That cuts both ways: nobody tracks when you log on, and you're also the one fixing your code on a Saturday when it's affecting users. The team already works like this; people diagnose an alert with Claude and ship the fix before anyone thinks to ask.
• End-user experience is our highest bar. Plenty of things are up for debate. The experience our users get is not.
• Flat, opinionated, and truth-seeking. There is no hierarchy to route ideas through: anyone can pitch anything and the team will genuinely listen. We will also push back hard to make sure it holds water, because we are truth-seeking above all else. Bring strong opinions and hold them loosely.
• Sustainable intensity. When it matters, we need you fully bought in. We also know that steering four AI agents at once gets exhausting fast and that rest is what keeps the work sharp, so we aim for a reasonable balance rather than burnout.

What Success Looks Like

In your first 90 days, you will have:
• Developed a deep mental model of the full AI pipeline and where its quality, cost, and latency actually live
• Shipped measurable improvements to auto-linking precision and recall, or to model quality
• Taken real ownership of the model serving and routing stack
• Shipped meaningful non-ML product and backend work alongside everyone else
• Formed a clear point of view on where we self-host versus call frontier models versus train our own
• Earned a reputation as a sharp, trusted reviewer of agent-generated code

Who You Are
• A systems thinker, above all. You operate at the architecture and design level, hold the whole system in your head, and treat code review as a first-class craft. You spot the solution that won't scale before it ships.
• Ambitious and fast-learning. We care far more about raw ability and drive than about pedigree. You're comfortable (or quick to get comfortable) steering AI tooling, and energized by it rather than threatened.
• Bonus: great taste and design sense. Not required, but a real plus; we sweat the details.
• Bonus: you love fashion. Also not required, but if Lekondo is a product you'd actually use, even better.

Ideal Background

We're genuinely flexible here. These are signals, not hard requirements:
• Hands-on experience somewhere in ML/AI: LLMs, computer vision, embeddings and retrieval, recommenders, or applied LLM work (prompting, evals, fine-tuning, serving)
• Strong backend engineering, ideally in a typed language at scale (we're in Go)
• Comfort with cloud infrastructure and MLOps (we run on AWS and GCP, with GPU clusters and Step Functions)
• Comfort in a fast-moving, ambiguous, early-stage environment

Location

For US candidates, this is an in-person role in New York City, five days a week. For non-US, remote.

Compensation

Competitive cash and meaningful early-stage equity. This is a founding-level role, so you'll own a real piece of what we're building, calibrated to experience.

How to Apply

Submit your resume along with your LinkedIn, GitHub, or portfolio, and a couple of sentences on why Lekondo. No cover letter required, and no application cap. We read every submission and will get back to you.