About the role
You will build LLM-powered agents for client workflows: support triage, research assistants, data agents. Beyond prompting, you learn the engineering that makes agents reliable: tool calling, retrieval, evaluation, guardrails and cost control.
Roles and responsibilities
- Design prompts and structured outputs that behave predictably
- Implement tool and function calling with safe execution
- Build retrieval (RAG) pipelines with embeddings and vector search
- Orchestrate single and multi-agent workflows
- Create evaluation sets and measure agent quality
- Add guardrails against prompt injection and bad outputs
- Track latency and cost and optimise both
By the end, you will be able to
- Build an agent that completes multi-step tasks with tools
- Ground answers in your own documents with citations
- Evaluate LLM systems instead of trusting vibes
- Ship an agent behind an API with logging and limits
- Speak confidently about agent architectures in interviews
Questions you will be able to answer
How does an AI assistant decide when to search the web instead of answering?
Why do chatbots "hallucinate", and how do teams stop it?
How can several AI agents work together without going in circles?
What stops someone from tricking an AI agent into leaking data?
What you might build
- A support-ticket triage agent
- A document Q&A assistant with citations
- A multi-agent research workflow with evaluations
Every intern gets different tickets. These are examples of the kind of work, not a fixed list.
How the first three months progress
Week 1
Onboarding & LLM fundamentals
LLM APIs (Gemini/OpenAI), tokens & context, prompting basics, API key hygiene
Week 4
Retrieval-augmented generation
embeddings, chunking, vector database, grounded answers with citations
Week 8
Evaluation
eval datasets, LLM-as-judge, regression tests, cost & latency tracking
Week 12
Capstone delivery
production agent for a business workflow, evals, docs, demo
Four- and six-month internships add advanced specialisation, a major client project, open-source work and interview preparation.
Tools you will use
Who should apply
- Comfortable with Python basics
- Have used ChatGPT or similar and want to know how it works underneath
- Enjoy experimenting and measuring results