- This is a role for a builder
- You won't just be managing Jira tickets; you will be writing production code, defining our alignment strategy, and shipping features used by millions of users globally
- Massive Scale: You will solve latency, throughput, and context management challenges at a scale of 80M+ daily tokens
- You will build sophisticated classifiers and alignment strategies to balance user freedom with strict safety guardrails
- Ship Code & Lead from the Front
- Act as a Player/Coach: Architect the system and mentor the team, but spend significant time hands-on in the codebase (Python/PyTorch)
- Own the core chat loop: Optimize context windows, memory/RAG retrieval, and inference latency to ensure a seamless, real-time experience
- Own the Model Lifecycle
- Drive our strategy for SFT (Supervised Fine-Tuning) and RLHF/DPO (Preference Optimization)
- You decide when to prompt, when to fine-tune, and when to architect a new RAG pipeline
- Manage the "Data Engine": oversee the sourcing, labeling, and cleaning of diverse datasets to improve model steerability and multicultural performance
- Architect High-Precision Moderation
- Build the immune system of the platform
- You will design and train custom classifiers to detect and filter non-consensual or illegal content within an explicit environment
Move beyond binary "safe/unsafe" flags to create nuanced, context-aware moderation systems
