AI for Survey Researchers – A three-workshop series (online)

Large language models are now embedded in research workflows across the social sciences, yet most researchers interact with these tools through consumer interfaces that obscure how they work, where data goes, and what decisions are being made on their behalf. This three-workshop series closes that gap. Across three standalone half-day sessions, participants build a working understanding of the AI stack: from how models generate text and where inference happens, through prompt engineering, retrieval-augmented generation, and API-based workflows, to the rapidly maturing ecosystem of agentic platforms, harness engineering, and autonomous research infrastructure. Each workshop combines conceptual exposition with live demonstrations and practical exercises grounded in survey research scenarios. No programming experience is required for Workshop 1; Workshops 2 and 3 assume familiarity with earlier concepts.

The course covers: 

• Workshop 1: How Large Language Models Work: tokens, training, alignment, data security, inference, open- vs closed-weights models, reproducibility challenges, and the limitations of chatbot interfaces for research.

• Workshop 2: Context Engineering: prompt design and optimisation, retrieval-augmented generation (RAG), API-based workflows and batch processing, memory and tool-calling, MCP servers, and evaluation engineering.

• Workshop 3: Agentic AI and Harness Engineering: the agentic AI ecosystem (IDE-native agents, extended-autonomy platforms, orchestration tools), harness engineering and SDKs, memory and token economics, MCP servers and hooks, oversight, auditability, and research transparency.

By the end of the course participants will:

• Explain how LLMs generate text and assess the implications of model architecture, training, and alignment for research practice
• Distinguish between open-weights and closed-weights models and evaluate their data governance implications
• Apply prompt optimisation techniques and build evaluation pipelines to validate LLM outputs
• Make structured API calls, manage parameters, and use retrieval-augmented generation where appropriate
• Map the agentic AI ecosystem, explain harness engineering, and assess how platforms orchestrate memory, tools, and context
• Design human-in-the-loop safeguards and audit protocols appropriate for agentic research workflows

Pre-requisites

No prior programming experience or specialist software knowledge is required for Workshop 1. Workshops 2 and 3 assume familiarity with concepts from Workshop 1 (or equivalent knowledge of how LLMs work). Workshop 3 benefits from some comfort with reading code, but participants are not required to write any. Setup guidance for API access will be provided before Workshops 2 and 3.

No software installation is required for Workshop 1. For Workshops 2 and 3, participants will benefit from having API access to a commercial LLM provider (e.g. Anthropic, OpenAI); setup guidance will be provided in advance. All demonstrations will be conducted live by the instructor. Participants do not need prior experience with any specific software, though basic familiarity with web browsers and text editors is assumed.

Target Audience

Survey researchers, methodologists, and quantitative social scientists across academia and government who use or are considering using large language models in their research. The series is designed to be accessible to researchers at all career stages, from doctoral students to senior investigators. 

Although the recommendation is to attend all three workshops, you are welcome to join them individually. 

PLEASE NOTE THESE WORKSHOPS WILL RUN ONLINE ON 23 NOVEMBER, 30 NOVEMBER AND 7 DECEMBER 2026 FROM 09:30-13:30.