Build AI that survives contact with production.
MediaStack builds AI products for newsrooms, brands, and creators. Our systems generate voices, recommend content, search decades of media, and occasionally remind us that a notebook working once is not the same as production.
We're looking for an engineer who enjoys experimenting, building complete systems, and shipping things people actually use.
Onsite · Tinkune, Kathmandu · Full-time
Build the AI layer across MediaStack's products.
- In-house text-to-speech models that sound natural and behave reliably in production.
- Content recommendation systems that help readers discover what matters to them.
- Smart archive discovery across decades of articles, images, audio, and video.
- New experiments and product ideas that are not on the roadmap yet.
You'll work across the full pipeline: research, data, models, APIs, deployment, monitoring, and improvements after real users inevitably find the edge cases. You'll have room to explore solutions, challenge assumptions, and take an idea from “this might work” to “people are using it.”
Models, infrastructure, and the usual production machinery.
Python and LLMs through Gemini, OpenRouter, and other providers.
In-house voice models, plus quantized and self-hosted models.
GPUs and frequent negotiations with VRAM.
Postgres, APIs, Docker, and Linear—because your head is not a project-management system.
You care about what happens after the demo works.
- You've built projects outside tutorials and coursework.
- Your GitHub contains experiments, side projects, unfinished ideas, or strange tools you made because they needed to exist.
- You care about retrieval quality, latency, model size, inference cost, and what happens after deployment.
- Reading about a new technique usually ends with you trying it.
- You enjoy taking ownership and figuring things out as you go.
- You communicate openly when something breaks or when you need help.
- You're comfortable working in a startup where products and priorities evolve quickly.
We care more about what you can build and how you think than degrees, titles, or years of experience.
Real ownership, real users, room to experiment.
Ownership over features used by real customers.
Direct collaboration with people building production AI systems.
Space to experiment, propose ideas, and turn good ideas into products.
Show us something you built. Polished, unfinished, useful, or weird.
Send us your GitHub, portfolio, project, experiment, or anything you think will interest us. It can be all four.
Apply at [email protected] →