About Raindrop
Raindrop is the monitoring platform for AI agents. Engineering teams at Fortune 100s and the fastest-growing AI companies (Vercel, Speak, Clay) use it to catch silent failures in production. When an agent misbehaves, Raindrop alerts the team, links to the trace, and surfaces the root cause so they can fix it fast. With Raindrop 2.0 (Self-Healing Agents), the loop closes itself: your coding agent fixes it, and the fix becomes an eval to prevent regressions.
Why It Matters
AI agents fail constantly in ways both hilarious and terrifying. Regular software throws exceptions. But AI agents fail silently, leaving engineers with almost no visibility into how their agents are actually performing.
The current status quo is sifting through millions of logs and trying to debug flaky evals that just aren't matching real world results. Evals are like unit tests, they confirm your model got specific test cases right. But in the real world agents call thousands of tools, run for hours, and encounter millions of unpredictable actions.
That’s where Raindrop comes in. It learns the unique shape of each AI agent’s issues. Starting from presets like Laziness, Forgetting, or Task Failure, to automatically tuning itself to each agent and discovering unknown unknowns.
With one click of a button, AI engineers start tracking issues or topics across 100% of their production data. They can see frequency over time, how many users are affected, relevant properties and more.
In order to process hundreds of millions of events, we gradually train small, custom models, private to each company, that learn to uniquely understand how their product is used.
As part of the early team, you’ll play a fundamental role in shaping the company - from making strategy and product decisions, to helping scale the team, to shaping the future of AI agents.
Our Investors
We've raised $50M total, led by CRV, with backing from Lightspeed, Figma Ventures, Vercel Ventures, and the founders of Replit (Amjad Masad, Michele Catasta), Cognition (Walden Yan), Framer (Koen Bok, Jorn van Dijk), Speak (Andrew Hsu), and Notion (Akshay Kothari).
About the role:
We’re looking for an ML Engineer to make Raindrop’s issue detection sharper, faster, and cheaper. You’ll own the models that find anomalies across hundreds of millions of agent events, from training through production, on an early team where your work reaches customers quickly.
Your Focus:
Anomaly detection: Improve how we find unusual agent behavior across every customer’s production data, including issues no one has defined yet.
Custom models: Train and serve the small, private models we build for each customer, and make them more accurate over time.
Labeling and feedback: Build the loops that turn customer signals into better models.
Cost and latency: Keep our models fast and affordable as event volume grows.
Your responsibilities:
Design, build, and scale the ML pipelines behind Raindrop’s issue detection.
Train, evaluate, and deploy models, and monitor their quality in production.
Iterate quickly without sacrificing quality.
Work with customers to understand how their agents fail and what they need to catch.
Bring new detection methods into production.
Ideal Candidate:
Ownership of ML systems end to end, from data and training through deployment and monitoring.
A track record of shipping models to production that real users depend on.
Experience scaling applications that serve millions of requests.
Expertise in an area such as anomaly detection, classification, representation learning, or LLM fine-tuning.
Good judgement about when to move fast and when to build for the long term.
Use and build AI agents and tools