Responsibilities
- Fine-tune large language models (LLMs) for style and tone alignment with The Economistโs editorial voice
- Design, curate, and manage datasets used for fine-tuning, including versioning and annotation workflows
- Build and evaluate RAG pipelines that incorporate retrieval from structured content
- Prototype and test TTS (text-to-speech) pipelines for use in audio-first products
- Collaborate with infra, frontend, and design leads to ship internal tools and demos that explore GenAI capabilities
- Own quality evaluation pipelines (BLEU, ROUGE, editorial scoring, custom evals), including human-in-the-loop feedback loops
- Support product experiments where real-time generation, summarisation, or personalization are being tested
- Partner directly with journalists and editors to develop novel evaluation metrics for editorial tone and style
Requirements
- 3+ years experience building with LLMs or NLP pipelines
- Experience with supervised fine-tuning, prompt tuning, or instruction tuning on proprietary datasets
- Strong understanding of fine-tuning paradigms, including RLHF/RLAIF preference modeling
- Strong Python skills including use of LangChain, HuggingFace Transformers, and data pipelines
- Comfortable defining and tracking generation quality with BLEU, ROUGE, and editorial-specific evaluators
- Exposure to STT / TTS tools for prototyping (Whisper, ElevenLabs, Bark, etc.)
- Strong communication and collaboration skills with editorial and product teams
- Curious, exploratory mindset comfortable with ambiguity and pushing GenAI boundaries
Preferred Qualifications
- Experience fine-tuning models for style or persona (chatbots with character voices)
- Exposure to observability tools for prompt / LLM testing (e.g., LangSmith)
- Knowledge of voice synthesis or emotion conditioning in audio pipelines
- Previous work in media, journalism, podcasting, or content production
- Interest or experience in multi-modal generation (text + image + audio)
- Contributions to open-source LLM tools or libraries
- Interest in ethical, editorial, and philosophical AI content considerations
Working Arrangements
- Full-time role based in London with a hybrid work pattern requiring 3+ days in office
