Most startups hire a content team. We're hiring the engineer who builds one out of software.

Recollia builds AI for medical students. Our marketing will be built the same way: a machine that turns one brief into a week of videos and carousels, in Mnemo's voice, in every language our students speak, published and measured without anyone opening an editor. You'll build it and own what it ships.

THE MACHINE

Monday, 9:00. A brief lands: "Cranial nerves, the mnemonic every second-year forgets." By 9:30 the system has made a 40-second video with Mnemo on screen, a 3-slide carousel, three alternate hooks, voiceovers in English, Hindi and Tamil, captions, thumbnails, and a schedule across Instagram, YouTube Shorts and LinkedIn. A doctor has checked the mnemonic. You approve it in one tap. By Friday you know which hook won, and next Monday's brief is already better.

That's what you're here to build. We have the tools and the audience. You bring the machine.

WHERE THIS GOES

This starts as an internship with a stipend of Rs 20,000 and a possible incentive of Rs 5,000 every two weeks, depending on performance. Get the machine to thirty pieces a week that students save, share and sign up from, at a cost per piece we'd be embarrassed to tell an agency, and it converts into a founding engineer role with equity. The bar is high on purpose: the internet is full of AI slop, and we're not adding to it.

WHAT YOU'LL BUILD

- Brief to script: Claude agents that research the topic and write the hook, the script and the carousel copy in our voice, with every medical claim sourced
- Script to visuals: Higgsfield workflows for images and video, Mnemo as a consistent character, avatars and lip-sync, templates so the look holds across a thousand pieces
- Script to voice: ElevenLabs for English, Sarvam for Hindi, Tamil, Telugu and more; one script, every language our students study in
- Assembly:



Remotion or FFmpeg pipelines that stitch, caption, brand and export to every platform's specs, untouched by hand
- Review: a one-tap approval queue with a doctor's check on the medicine and a brand check on the look; every rejection teaches the system
- Publish and measure: scheduled posting through the platform APIs, every asset traced from view to save to signup, and a weekly loop that feeds the winners back into next week's briefs
- Launches on autopilot: every feature we ship gets its announcement video, carousel and post generated from the release notes

REQUIRED

How it's done

- You've built this before in some form: an unattended pipeline that turns inputs into published content, and you can show it running
- Claude agents at production depth: tool calling, structured outputs, MCP servers, long-running loops that fail protected and retry
- Orchestration and cost: queues, idempotency, cron; n8n or Make when it's enough and code when it isn't; you know the cost of every piece to the rupee The tools
- Higgsfield: image and video generation, character consistency, avatars, lip-sync, motion control, and the prompt and workflow craft behind a look that holds
- ElevenLabs: voice design, cloning, multilingual speech, dubbing
- Sarvam: Indian-language speech and text, and where it beats the global models for our students
- Claude: research, scripts, hooks, copy, and the agent that runs the show
- Remotion, FFmpeg or equivalent for templated assembly; platform APIs for Instagram, YouTube, LinkedIn and X,



and schedulers for when an API says no

Taste and craft

- You can tell AI slop from a scroll-stopper in one second, and you refuse to ship the first kind
- Hooks and retention: why someone stops in the first second and stays to the last
- Carousels that get saved, Shorts that get shared; you know every platform's specs and quirks by heart
- Copy for medical students: their exams, their slang, their 1 a.m. panic, without condescension
- A brand system in code: templates, tokens, fonts, Mnemo's rules

Measurement

- Saves, shares, follows and signups traced back to each asset: UTMs, PostHog or equivalent
- Ten hooks, not one; the numbers choose, and the winners become next week's templates
- Medical accuracy as a hard gate: every claim sourced; a wrong fact in a student's feed is worse than no post

Engineering

- Python or TypeScript at production level: async, typing, tests
- FastAPI or equivalent, Postgres, S3, Docker, Git, CI
- Media handling: codecs, formats, and file sizes that upload the first time

No degree required. Skills over credentials.

WHAT HUNGER LOOKS LIKE TO US

- You've posted every day for months with a machine you built, and you can show the numbers.
- You'd rather ship ten variants than argue about one.
- A piece that looks like every other AI post bothers you until it's fixed.
- You ask for more scope, not more instructions.
- You say when something is broken, including our decisions.

HOW TO APPLY

- Email [email protected] with the subject "Content Machine" and one pipeline you built: repo, demo, or write-up. In three sentences, tell us what it does, what broke, and what you'd do differently. Then make us one 30-second video on any topic a medical student would stop scrolling for, with whatever AI tools you like, and send the link with two numbers: how long it took you, and what it cost.

📌 AI Content Automation Engineer (Founding Track) (India)
🏢 Recollia
📍 India