About Codeium
We're one of the fastest growing AI startups, focused on product, revenue, and customer experience. We work hard, and we operate with a high degree of trust, agency, and ownership.
What you'll do
- Train and fine-tune large language models
- Navigate high levels of uncertainty and prioritize high-value ML experiments to maximize product impact
- Demonstrate initiative and the ability to start and make progress on projects independently
- Swiftly design, track, and analyze experiments results. Meticulously document findings, conduct ablation studies, and synthesize data into actionable insights.
- Participate in the ML reading group and level up the team's knowledge of LLM training and infrastructure
About you
- Strong software engineering skills. There are no pure research scientists at the company.
- Strong grasp of the feasibility frontier of CS, AI, and LLMs, from H100 bandwidth to GPT-4 capabilities to vector database performance.
- Deep curiosity about the code generation problem. Willingness to constantly re-examine priors in the face of new discoveries.
- Skilled in transforming successful experimental outcomes into robust, scalable features for the core product offering
- Experience training and iterating on large production neural networks in any domain (self-driving, language models, etc.) is a strong plus
- Familiarity with AI-powered developer tools like Codeium, Copilot, ChatGPT, and others is a strong plus
What we believe
- Our best work is done in person. The team goes in 5 days a week into our office in downtown Mountain View, CA (within walking distance of the Caltrain station).
- Research is in service of a better product. While we read many papers, we won't have time to write them. The best AI researchers have excellent software engineering skills and know that infrastructure and evaluation work are critical.
Recent projects
Some of the things that our research-focused software engineers have worked on recently
- Regularly deploying an autocomplete and chat product that scales to hundreds of thousands of daily active users.
- Instruction and edit fine-tuned models for Codeium Command.
- Realtime context retrieval
- Codebase fine-tuning