AI Engineer

Description

365scores is seeking a self-driven AI Engineer to join our Core AI team. You will own the full product lifecycle of end-to-end generative AI systems—from data extraction and backend database design to multi-agent orchestration, evaluation, and live production monitoring.

Your domain will bridge two high-stakes engineering environments

high-throughput batch processing for deep pre-match analysis and ultra-low-latency generation (sub-second) for live match streaming. You will operate with high autonomy to drive architectural decisions and shape our applied AI infrastructure.

Responsibilities

• Multi-Agent Orchestration & Systems: Design, build, and deploy specialized LLM agents for data triage, real-time content generation, multilingual translation, and automated self-critique using agentic frameworks (e.g., Google ADK).
• Harness & Memory Engineering: Build agent harness frameworks, manage stateful session handling, and implement scalable short-term and long-term memory architectures.
• Real-Time & Batch Workflows: Architect dual-track AI workflows: optimize pre-match pipelines for analytical depth, and tune live match generation pipelines for sub-second response times.
• Model Routing & Cost Optimization: Implement intelligent dynamic routing between commercial and open-source models (via OpenRouter, AWS, GCP, and Databricks) to balance latency, cost, and quality.
• Data Engineering & RAG: Design robust database schemas and data pipelines that fuel production RAG architectures and dynamic agent context.
• Evaluation & Observability: Build automated critique loops (LLM-as-a-judge) to measure output precision, while implementing full observability/logging to trace agent reasoning steps in production.

Requirements

• 3+ years as a software engineer, ML engineer, or data scientist, with strong Python and a track record of code running in production
• Solid backend and database design
• At least one LLM-powered feature shipped to production, using commercial or open-source models. You can describe what broke and how you fixed it
• Hands-on experience post-training open-source models (e.g., SFT or LoRA), including preparing training data and measuring whether the result actually beat the baseline
• Experience evaluating non-deterministic outputs, even informally
• Comfort with latency-sensitive Python (async, streaming, queues, caching)
• Hands-on cloud experience (AWS, GCP, or Databricks)
• Comfort working with product and data partners to turn loose ideas into shipped features
• Disciplined use of AI coding tools: choosing the right model per task, giving structured context, and verifying output rather than “vibe coding”
• Curiosity backed by evidence: you benchmark new tools before adopting them

Nice to have

• Multi-agent or agentic systems in production
• Experience with prompt caching, managed LLM endpoints, or model routing
• Real-time event streaming for sports or live media
• Multilingual generation or translation quality evaluation
• Open-source contributions to agentic tools or eval benchmarks
• Sport fanatic - Advantage