Skip to main content
TALENT247 // SPECIALIST AI EXECUTIVE SEARCH & TECHNICAL RECRUITMENT
ML · Machine Learning

Senior Spoken Dialogue & Small Language Model Engineer

London, UK · Full-time · Hybrid

AI Specialism

Machine Learning

Workplace

Hybrid

Location

London, UK

Compensation

£95k – £135k

annual package

Role Dossier & Specification

REF: SENIOR-SPOKEN-DIAL

THE CLIENT & MANDATE

Our client, spun out of Cambridge's renowned spoken-language systems research lab and headquartered in London, is Europe's category leader in enterprise conversational voice agents. Serving hundreds of global brands across banking, insurance, healthcare, and hospitality, their platform must resolve complex multi-turn spoken transactions—authentication, alphanumeric account codes, interruptions, regional accents, and noisy mobile lines—within strict sub-500ms conversational response budgets.

They have partnered with Talent247 to appoint a Senior Spoken Dialogue & Small Language Model Engineer in London. Sitting inside their Core Dialogue & Modelling R&D group, you will post-train specialized small language models (SLMs) and joint speech-text encoders engineered specifically for spoken turn-taking, zero-hallucination tool execution, and robust slot extraction over imperfect ASR hypotheses.

KEY RESPONSIBILITIES

  • Post-train and distil compact 1B–14B spoken-dialogue language models using SFT, DPO, and online RL from simulated caller rollouts to achieve frontier-level reasoning inside sub-150ms token-generation budgets.
  • Design ASR-error-resilient architectures—consuming N-best hypotheses, phoneme/acoustic embeddings, and prosodic turn-end cues—to eliminate premature interruptions and misheard alphanumeric entities (postcodes, policy IDs, dates).
  • Architect constrained tool-calling and schema-guided decoding layers guaranteeing that live voice agents never fabricate enterprise policies, pricing, or account actions.
  • Build multi-turn conversational simulation environments where adversarial synthetic user agents test complex clarification, repair, and escalation strategies across dozens of languages and dialects.
  • Collaborate closely with GPU inference engineers to quantize and serve multi-LoRA tenant adapters with high prefix-cache efficiency on shared vLLM/TensorRT-LLM clusters.
  • Track and publish applied advances in spoken dialogue systems, turn-taking modeling, and task-oriented LLM alignment.

TECHNICAL ENVIRONMENT & OFFERING

Their R&D stack spans Python, PyTorch, FSDP/DeepSpeed, TRL/verl, vLLM/TensorRT-LLM, streaming speech encoders, and multi-GPU AWS/dedicated clusters. Working hybrid in West London, our client offers £95,000–£135,000 GBP base salary plus generous Series D stock options, private medical insurance, pension match, and UK Skilled Worker / Global Talent visa support.

Core Requirements

ESSENTIAL QUALIFICATIONS

  • 5+ years of experience in NLP, conversational AI, speech-language modeling, or LLM post-training engineering.
  • Deep hands-on proficiency in PyTorch, transformer fine-tuning/distillation (SFT, LoRA, DPO, RLHF/GRPO), and low-latency small-language-model (SLM) optimization.
  • Strong understanding of spoken-language phenomena—ASR noise robustness, endpointing/turn-taking, disfluencies, coreference, and multi-turn dialogue state tracking.
  • Proven track record building rigorous offline and simulated multi-turn evaluation suites that correlate reliably with live production call containment and CSAT.
  • Clean software engineering habits in a shared Python/C++ research-to-production codebase.

PREFERRED / BONUS EXPERIENCE

  • PhD or MSc in Spoken Dialogue Systems, Computational Linguistics, Speech Recognition, or Machine Learning (Interspeech, ICASSP, ACL, EMNLP publications a plus).
  • Experience with joint audio-text multimodal LLMs or speculative/grammar-constrained decoding in vLLM/TensorRT-LLM.

Position Telemetry

Location London, UK (Hybrid)
Department Machine Learning
Employment type Full-time
Workplace type Hybrid
Salary range
£95k – £135k annual package