Principal LLM Engineer

ReSpo.Vision is an AI and Computer Vision company transforming how sports are analyzed, visualized, and monetized. Our proprietary single-camera system extracts elite-level tracking data and performance analytics from standard broadcast or tactical video, without wearables or in-venue installations. Already used by global clubs, federations like FIFA, Concacaf, or CONMEBOL and for competitions like Euro or Copa America, we are actively expanding into media, fan engagement, and betting applications.

Our pipeline combines advanced computer vision and deep learning models to track all players and the ball in 3D using a single-camera feed. The resulting positional data powers our growing product suite: from match analytics platform to visual content, including 3D match reconstructions (see an early demo: YouTube) and real-time broadcast augmentation with dynamic stats and virtual overlays. The system is built for scalability, leveraging cloud-native infrastructure, GPU inference pipelines, and sports-specific post-processing modules that turn raw detections into meaningful insights.

We’re now entering an exciting phase, pushing our tracking system toward real-time applications, lowering latency to enable live insights and instant visual augmentations. In parallel, we’re building new layers on top of our core data, including realistic 3D match reconstructions and virtual overlays for enhanced broadcast experiences.

Your Role

As a Principal ML Engineer specializing in LLMs at ReSpo.Vision, you'll build intelligent systems that transform how we understand and narrate sports. This is a hands-on position where you'll integrate cutting-edge language models with our sports tracking data, implement production LLM pipelines, and create AI-powered features that enhance our analytics and content generation capabilities. You'll own LLM-based initiatives from conception through deployment, personally coding at the intersection of NLP, sports data, and real-time systems.

You will be responsible for:

1. LLM System Architecture: Design and implement production LLM pipelines that process our tracking data into natural language insights, automated commentary, and intelligent analytics
2. Sports Intelligence Platform: Build systems that combine our computer vision outputs with LLMs to generate match summaries, player insights, tactical analysis, and real-time commentary
3. Retrieval & Context Systems: Develop RAG architectures that leverage our vast sports database, enabling LLMs to provide accurate, context-aware responses about matches, players, and tactics
4. Multi-Modal Integration: Connect our visual tracking data with language models - turning player movements into narratives, tactical patterns into insights, and match events into engaging content
5. Real-Time LLM Infrastructure: Build low-latency systems for live match narration, instant stat explanations, and interactive fan experiences powered by LLMs
6. Prompt Engineering & Fine-tuning: Optimize LLMs for sports-specific tasks through prompt engineering, fine-tuning, and potentially training sport-specific models
7. API & Integration Layer: Create robust APIs that allow our clients to leverage LLM capabilities, from automated match reports to conversational sports analytics

Who You Are

1. You've spent 5+ years building ML systems in production, with at least a year focused on LLMs and NLP
2. You've implemented RAG systems or fine-tuned language models, and built production LLM pipelines
3. You understand the full LLM stack: from embeddings and vector databases to inference optimization and prompt engineering
4. You can design systems that handle both accuracy (sports facts matter) and creativity (engaging narratives)
5. You enjoy connecting different data modalities - turning numbers, positions, and events into meaningful language
6. You're comfortable with the full stack: data pipelines, model serving, API design, and monitoring LLMs in production
7. You understand the challenges of LLMs at scale: cost optimization, latency, hallucination prevention
8. You're excited about sports and see the potential for AI to transform sports storytelling and analysis

Core Technical Requirements

1. Deep experience building production LLM systems - RAG, prompt optimization, optionally fine-tuning
2. Strong understanding of modern LLM architectures and frameworks (Transformers, LangChain, LlamaIndex)
3. Experience with vector databases, embedding systems, and semantic search
4. Track record of shipping LLM products that handle production scale and reliability requirements
5. Strong software engineering fundamentals and API design skills
6. Mastery of LLM-based development tools and ability to leverage them effectively in daily work
7. Experience building real-time or streaming LLM applications

Nice to have

1. Experience with multi-modal AI systems that combine text with structured data
2. Experience with sports data, commentary systems, or content generation
3. Experience training or fine-tuning large language models from scratch
4. Contributions to open-source NLP/LLM projects
5. Expertise in LLM inference optimization - quantization, caching, batching strategies
6. Experience with structured data-to-text generation systems

What we offer

1. A chance to work with a top-tier engineering team, including Kaggle Grandmasters
2. Hybrid work model
3. Flexibility in employment type (B2B/contract of employment)
4. Market-competitive salary
5. Private healthcare and Multisport card
6. Open training budget – we’ll support your development in relevant areas
7. Ownership and autonomy – no micromanagement, real impact
8. A unique opportunity to shape a globally recognized, high-impact product used by top sports organizations like Chelsea, Paris Saint-Germain, or FIFA