Intern (TGI - Machine Learning/ AI Engineering)

What matters most

• Applied Machine Learning Research

  The role requires researching and implementing automated feature engineering, feature selection, and model optimization techniques. You must be able to translate these research concepts into practical, reliable systems for the Templeton Global Investments team.

• Software Engineering for AI Pipelines

  Beyond modeling, you must develop reproducible machine-learning pipelines and build proof-of-concept AI tools. This requires writing tested, maintainable, and well-documented Python code, as well as building backend services and APIs.

• Stakeholder Communication and Documentation

  You are expected to track experiments, evaluate computational performance, and present findings to stakeholders. Clear communication of model limitations and recommendations is essential for the successful adoption of your AI tools.

Likely challenges to think through

• Bridging Research and Production

  A common challenge is the gap between experimental data science and production-ready software. Prepare by focusing on writing modular, version-controlled code using Git and understanding how to wrap models in REST APIs or lightweight interfaces.

• Domain-Specific Modeling Complexity

  The role emphasizes time-series modeling and optimization methods like genetic algorithms. Prepare by reviewing how these specific techniques are applied to financial or quantitative data, ensuring you can validate results critically in a professional context.

Topics to prepare

• Machine Learning Optimization and Feature Engineering

  The role requires researching automated feature engineering and model optimization. Exercise: Implement a pipeline using scikit-learn to compare different feature selection methods on a time-series dataset, documenting the impact on model performance metrics.

• Time-Series Modelling and Validation

  The position involves working with financial data, which often requires time-series analysis. Exercise: Build a validation framework that uses walk-forward cross-validation instead of standard k-fold to prevent data leakage when evaluating a time-series forecasting model.

• Agentic Workflows and RAG

  The role values interest in LLMs and agentic workflows. Exercise: Create a simple proof-of-concept using a framework like LangChain to implement a RAG pipeline that retrieves information from a local document set to answer specific queries.

Skills to revise

• Production-Oriented Python Development

  The role emphasizes writing maintainable, tested code. Exercise: Refactor a monolithic script into a modular Python package with unit tests using pytest, ensuring all functions are documented with docstrings.

• Data Pipeline and SQL Integration

  The role requires building reproducible pipelines. Exercise: Write a SQL query to extract and aggregate financial data, then create a Python script that cleans this data and saves it into a structured format suitable for model training.

• Git and Experiment Tracking

  The role requires tracking experiments and working with existing codebases. Exercise: Use Git to manage a project with multiple branches for different model experiments, and use a tool like MLflow to log parameters and metrics for each run.

• Build a reproducible machine learning pipeline

  Create a project that demonstrates a complete end-to-end workflow using Python, pandas, and scikit-learn. Ensure your code is modular, well-documented, and version-controlled using Git, as the role emphasizes building reusable components and maintainable software.

• Demonstrate proficiency in model optimization

  Prepare a technical walkthrough of a project where you applied hyperparameter optimization or feature selection techniques. Be ready to discuss how you evaluated model performance and tracked experiments, specifically using libraries like XGBoost or custom optimization methods.

• Develop a proof-of-concept AI tool

  Build a small-scale application or backend service that incorporates a REST API. If possible, integrate a RAG (Retrieval-Augmented Generation) workflow or an agentic concept to showcase your interest in modern LLM applications and your ability to build lightweight interfaces.

• Review time-series modelling and genetic algorithms

  Refresh your theoretical knowledge on time-series validation and evolutionary computation. Be prepared to explain how these methods can be applied to financial data or asset management problems, as these are specific areas of interest mentioned in the job description.

• Prepare for technical communication

  Practice summarizing complex technical findings for a non-technical audience. Prepare a brief presentation or documentation sample that outlines a problem, your methodology, the limitations of your approach, and your final recommendations.

• Formulate questions for the team

  Prepare thoughtful questions regarding the team's current data infrastructure, such as how they manage CI/CD for machine learning models or how they balance experimental research with production-level software engineering requirements.