Lead AI Engineer

AI Fund is a mission-driven startup reinventing the way people find peace and inspiration through digital experiences. The Lead AI Engineer will architect and evolve the cognitive engine of the platform, focusing on model tuning, inference optimization, and intelligent orchestration.

Responsibilities:

• Lead fine-tuning of foundational models using efficient training techniques and custom datasets
• Design and implement model orchestration logic that determines when to retrieve, route, generate, or escalate across different conversational contexts
• Build and iterate on eval frameworks for long-form, multi-turn interactions - prioritizing emotional coherence and user outcomes over token accuracy
• Stay on top of rapid developments in LLMs, fine-tuning frameworks, and inference efficiency, translating that knowledge into action
• Champion best practices for scaling training workflows, experimenting safely, and continuously learning from real-world feedback
• Design, iterate, and evaluate prompt strategies for complex multi-turn interactions using frameworks like DSPy
• Build prompt libraries and A/B test variants to optimize for safety, clarity, and on-brand responsiveness
• Leverage prompt engineering as a short-term strategy where fine-tuning is not yet appropriate, with a clear view on trade-off
• Evaluate and integrate modular orchestration strategies (e.g., LangGraph, LlamaIndex, Letta, PydanticAI), forming a perspective on their relevance and scalability
• Design systems that can switch between reflection, coaching, or directive states based on context, using either routing logic or learned behavior
• Collaborate with the product team to define how tools, memory, and reasoning modules interact without overcomplicating the user experience
• Own parameter-efficient fine-tuning pipelines (e.g., LoRA, QLoRA) to adapt foundational models to brand-specific voice, tone, and emotional range
• Curate high-quality datasets and design eval metrics tailored to coherence, empathy, and state consistency across sessions
• Explore model compression, quantization, and inference optimization for low-latency voice and mobile interactions
• Design lightweight experiments to validate technical approaches and measure outcomes beyond accuracy (e.g., trust, emotional congruence)
• Partner with domain experts to implement human-in-the-loop annotation systems where automation falls short
• Ship prototypes and production features rapidly, with a build-learn-refine approach