Senior Applied ML Engineer, Evals & Data

Cardboard is an agentic video editor for growth/marketing teams and serious creators who need to ship videos consistently. Instead of agencies, long timelines, and infinite review cycles, teams give cardboard a footage + a goal (“3 variants”, “30s hook”, “testimonials”) to get a strong first cut in minutes on which they can iterate quickly.

Cardboard believes AI video creation should make creators faster without taking the creative control out of their hands.

Its editor brings generation, editing, and export into one place, and learns how each creator works along the way.

As Senior Applied ML Engineer at Cardboard, you'll own the quality system behind its AI video agent.

Your job is to turn subjective agent behavior into measurable signals, then make it better through datasets, evals, graders, and review. It takes real applied ML depth and a track record of shipping agent systems to actual users.

You'll work directly with the founders and product team.

• The agent quality bar. Define measurable standards for useful, reliable behavior across the editing workflow.
• Evaluation systems. Build offline and online evals, model graders, human review, and automated checks.
• Production datasets. Turn real usage and user telemetry into datasets for model and product iteration.
• Multimodal measurement. Evaluate search across video, audio, images, and graphics under real-time constraints.
• Regression gates. Set release checks that protect quality while tracking latency and cost.
• Model selection. Choose the models behind the agent and measure their production tradeoffs.
• Learning loops. Evaluate next-action prediction, creator memory, brand fit, pacing, and technical quality.

• Shipped an LLM agent to customers. You owned production quality after users met the system.
• Built evals for agent behavior. You translated broad quality concerns into metrics and test suites.
• Created datasets from real usage. You turned telemetry and human feedback into reliable training signals.
• Balanced quality, latency, and cost. You made model choices against production constraints.
• Measured subjective outputs. You designed review methods for preferences, style, or creative quality.
• Worked across research and product. You converted uncertain findings into systems that shipped.
• Built strongly in Python or TypeScript. Your implementation quality supported repeated model iteration.

• Turn uncertainty into a test. Define the signal, dataset, and threshold before arguing about model quality.
• Start with real user behavior. Build quality systems from production usage and direct feedback.
• Protect quality at release. Use regression gates to catch losses before updates reach customers.
• Trade off with evidence. Measure quality beside latency and cost when choosing models.
• Ship every week. Keep experiments small enough to learn and improve the product continuously.
• Work across functions directly. Partner with founders, product, and engineering without added layers.
• Care about the final edit. Treat technical quality, pacing, and brand fit as product outcomes.

Base pay is ₹60 lakhs a year. Equity ranges from 0.1% to 0.5%, plus founding ESOPs. You will work directly with the founders and own the agent quality system. The role gives you authority over how an AI video editor measures and improves intelligence.