AI/ML Engineer, only

About the Role

Xuman.Ai is building a network for on-demand human connection. The product is built, live, and people are using it. This role focuses on syncing user feedback with product development through a cycle of shipping, watching, learning, fixing, and improving.

What You'll Own

• The AI that greets, talks, and acts for a user within a live app on both app stores.
• The agent layer: including tool calling, orchestration, memory, routing between models, and handling failures or slowdowns.
• Voice and real-time pipelines: processing speech input, reasoning, and generating speech output within latency constraints.
• Safety and content: implementing checks for actions and content reaching users, ensuring bios, captions, and insights sound human-like.
• The ML foundation: managing captured signals, underlying data, evaluations for every change, tracking cost per conversation, and developing initial models when data permits.

How We Work

• The founder defines product direction and feel.
• Gaps, bugs, or improvements are identified and discussed before building.
• Once agreed, the technical approach is owned by the engineer.

Requirements & Tech Stack

• Experience building with LLMs and shipping agentic applications to real users.
• Production experience with agents and tool calling, including handling tool failures or model performance issues.
• Designed and implemented multi-step or multi-agent orchestration.
• Experience with LangGraph or equivalent, with a focus on simplicity.
• Practical machine learning experience: training, fine-tuning, or evaluating models, with the ability to interpret evaluation metrics.
• Experience building evaluations to prove changes.
• Proficiency in Python for production systems.
• Comfort working with TypeScript services (NestJS gateway is in use).
• Daily use of structured outputs, streaming, and function calling across multiple model providers.
• Treating model cost and latency as engineering problems.
• Actively following advancements in the field and incorporating new learnings into existing work.
• Daily use of Claude, Copilot, or ChatGPT.

What Helps Further

• Real-time voice on LiveKit.
• Effective retrieval implementation.
• Production experience with safety and moderation systems.
• Production experience with recommendation or matching systems.
• Experience with data pipelines at significant volume.

What You Get

• End-to-end ownership of your work.
• Work deployed to real users on a live product within the same week.
• Pay that increases with milestone achievements.
• Growth opportunities tied to performance and responsibility.
• Opportunity to join at an early stage and build the foundation, with foundational contributions recognized.