• Architect and evolve our multi-agent orchestration platform (currently built on Hermes / Multica), including plugin systems, tool-use pipelines, observability hooks, and channel adapters (voice, telephony, messaging)

• Design and implement voice AI pipelines — STT (VibeVoice-ASR, Whisper), real-time TTS with streaming (VibeVoice-Realtime), VAD (Silero), SIP/RTP telephony integration — with sub-300 ms end-to-end latency targets

• Build and maintain RAG pipelines with retrieval quality measurement, re-ranking, and hybrid search over vector + keyword indexes

• Define MCP server architecture and tool-use contracts across internal and third-party integrations

Model Development & Fine-Tuning

• Fine-tune and evaluate LLMs (LoRA, QLoRA, DPO) for domain-specific tasks including customer support, classification, and structured extraction

• Evaluate and benchmark model quality using automated evals, human preference data, and domain-specific metrics (WER, DER, cpWER for speech; RAGAS / LLM-as-judge for RAG)

• Manage model lifecycle: experiment tracking, versioning, reproducibility, and promotion to production

Observability & Reliability

• Own the AI observability stack: Langfuse tracing, span-level LLM call instrumentation, cost tracking, and quality regression alerting

• Define and enforce guardrails: hallucination detection, PII redaction, output safety scanning, and rate-limiting across multi-tenant deployments

Platform & Pipelines

• Build data ingestion, preprocessing, and feature pipelines supporting model training and continual learning

• Drive CI/CD for ML: automated eval gating, shadow deployments, canary releases, and rollback triggers

Technical Leadership

• Set architectural standards for AI systems across the group; conduct design reviews and own ADRs for major decisions

• Mentor ML engineers and applied scientists; grow the team's capabilities in production AI, not just prototype AI

• Collaborate with Product and Commercial teams to translate business problems into ML problem formulations with clear success metrics

• Engage with external research partners and track emerging work (arXiv, conference proceedings, open-source releases) to identify signals worth productionizing

Experience & Skills:

• 8+ years in ML Engineering, Applied AI, or Research Engineering with at least 2 years in a lead or staff-level role

• Deep, hands-on experience with LLMs in production: fine-tuning, RLHF/DPO, prompt engineering, RAG, and tool use

• Fluent in Python and the core ML stack: PyTorch, Transformers (HuggingFace), PEFT/LoRA

• Real experience with LLM inference serving — vLLM, TensorRT-LLM, or TGI — in a latency-sensitive production environment

• Practical knowledge of agentic frameworks: multi-agent coordination, tool-call orchestration, context/memory management, and observability (Langfuse, Opik, or equivalent)

• Experience with speech AI (ASR/TTS pipelines) or real-time audio systems is a strong plus

• Solid understanding of MLOps: experiment tracking (MLflow/W&B), model registries, containerization (Docker/Kubernetes), and CI/CD for ML

• Awareness of LLM-specific risk: hallucination, prompt injection, data leakage, fairness, and privacy — and how to mitigate them in production

• Strong communication skills: you can write a crisp design doc, run a productive architecture review, and explain tradeoffs to a non-technical stakeholder

Nice to have

• Experience with voice pipelines end-to-end: VAD → ASR → LLM → TTS → SIP/RTP telephony

• Multi-hop RAG with self-consistency, chain-of-thought reranking, or RAPTOR-style hierarchical retrieval

• Familiarity with MCP (Model Context Protocol) server design and tool-use contracts

• Contributions to open-source ML projects or published work (arXiv, NeurIPS, ACL, Interspeech, etc.)

• Experience with multimodal models (vision-language, audio-language)

• Knowledge of quantization techniques (GPTQ, AWQ, GGUF) and their quality/latency tradeoffs

Right to Work

At any stage, please be prepared to provide proof of eligibility to work in the
country you’re applying for. Unfortunately, we are unable to support relocation
packages or sponsorship visas

"Come as you are"

Everyone is welcome here. Diversity & Inclusion are at our core. Far above any technical competence, we value respect, openness, and trusted collaboration. We do not tolerate intolerance.

ESG

"At team.blue, our commitment to caring for the environment and each other is at the heart of everything we do. Our latest impact report showcases our ongoing ESG efforts and ambitious sustainability goals. Interested in learning more about our dedication to making a positive impact? Check it out here.”

#LI-CC1