About

About the Project We’re developing an internal AI assistant designed to provide accurate, evidence-based, and brand-consistent responses to public and internal questions about complex or technical topics. The tool will function as a knowledge-grounded answer engine capable of retrieving and summarising verified information from trusted documents and then expressing it in a defined tone of voice. It will eventually support two distinct brand personas, each with its own tone and regional spelling conventions. Initially, it will be used internally by our communications team, with potential to expand into a client-facing or public tool. What We’re Building A retrieval-augmented generation (RAG) system that can: Ingest and index PDFs, reports, FAQs, and internal knowledge documents. Retrieve and summarise relevant information to answer user questions. Write replies in two distinct brand tones, each with unique personality and language rules (e.g., US vs UK English). Include citations, jurisdiction tagging, and clear source attribution. Decline or flag questions where evidence is weak, conflicting, or absent. Tone and accuracy are equally important — the assistant should sound human, credible, and trustworthy at all times. Core Responsibilities Build the prototype of a tone-aware RAG chatbot using modern LLM frameworks (OpenAI, Anthropic, etc.). Design and implement a knowledge ingestion pipeline for structured and unstructured sources. Create a tone-conditioning layer (prompt-based or fine-tuned) for two distinct writing styles. Implement guardrails, citation logic, and tone validation to maintain reliability. Develop a simple web or Slack interface for internal testing and feedback. Produce documentation for updating data sources and refining tone behaviour. Required Skills Proven experience developing LLM-based applications (RAG, embeddings, or fine-tuning). Strong Python skills, ideally with frameworks like LangChain, LlamaIndex, or Haystack. Familiarity with vector databases (Pinecone, Weaviate, FAISS, etc.). Understanding of prompt engineering and tone control in natural-language generation. Experience implementing content moderation, retrieval filters, and citation validation. Excellent written communication and clear documentation practices. Bonus Points For Experience building AI systems for science communication, media, or brand content. Work on multi-tone or multi-brand LLM models. Knowledge of UX for chat or Q&A interfaces. Deliverables Functional prototype of the assistant (MVP). Documentation outlining ingestion, retrieval, tone, and guardrail mechanisms. Optional: architecture plan for scaling to external/public deployment. How to Apply Please include: - A brief summary of your experience building AI or RAG systems. - One or two project examples (screenshots, demos, or GitHub links). - A short note describing how you’d technically approach dual-tone, evidence-grounded responses. - Timeline and cost estimate