AI Engineering Intern

Responsibilities

Work with Trexo Global’s engineering team to improve the accuracy and speed of opensource large language models (LLMs) hosted on AWS. You will test changes against real intellectual property workflows.

Improve model accuracy - Compare model responses with expected answers, identify recurring errors and test changes that address them.

Build an evaluation harness - Write Python tests that compare the current model with a proposed change using the same test cases. Measure accuracy and response time on a held-out test set.

Test changes to the model pipeline - Vary prompts, document chunking and output schemas one at a time. Record which changes help and turn recurring failures into regression tests.

Build an AI agent and its harness - Prototype an agent that selects a tool and checks its result. Validate tool arguments, handle timeouts and set a limit on retries.

Integrate tools through MCP - Use a Model Context Protocol (MCP) client to discover and call tools exposed by an MCP server. Test what happens when a lookup returns no match or the tool is unavailable.

Benchmark inference on AWS - With an engineer, test how context length and concurrent requests affect latency, throughput and GPU memory on an AWS EC2 instance. Check response quality before recommending a configuration.

Make experiments reproducible - Use approved, sanitised sample documents. Record the model version, settings and results in Git so another engineer can repeat the experiment.

Qualifications

Education - Pursuing or recently completed a degree in Computer Science, Engineering or a related field, or equivalent practical experience.

Project experience - Built a small application, course project or personal experiment that calls an LLM.

Programming - Comfortable writing Python, calling REST APIs, working with JSON and using Git. Able to write a test that checks an output against an expected result.

LLM fundamentals - Understand prompts, tokens, context windows and structured outputs. Able to explain how you would check whether a model’s answer is supported by the source document.

AI agents and MCP - Understand how an agent chooses a tool, how a harness manages execution, and how an MCP client calls a tool on a server.

Preferred Skills

Model serving - Exposure to Linux, Docker or AWS EC2, or experience running an open-source model locally with llama.cpp or a similar inference server.

Model adaptation - Coursework or experimentation with retrieval-augmented generation (RAG), quantisation or parameter-efficient fine-tuning such as LoRA.

Evaluation datasets - Experience building a small labelled dataset or reviewing model responses against expected answers.

Soft Skills

- Investigate unexpected results patiently and distinguish a reproducible improvement from a promising one-off example.

- Explain findings clearly, ask for feedback and work with engineers and domain reviewers to resolve ambiguous cases.