NLP / AI Engineer

What matters most

• Combine textual datasets

  The role explores and combines internal and external text for NLP research. Be ready to discuss how you would prepare and analyze text from different sources using Python and SQL.

• Build generative AI solutions

  The posting calls for solutions using both internal models and external APIs, and cites experience with agents, RAG, and fine-tuning and deploying open-source models. Focus on when each approach serves a task.

• Formulate investment research hypotheses

  The team aims to derive investment alpha from NLP research, with potential portfolio impact. Prepare to explain how a text-based signal could become a testable research hypothesis, without assuming it will improve returns.

• Post-train and benchmark for internal tasks

  The role explicitly includes task-specific post-training and benchmarking LLM approaches on internal tasks. Be ready to explain evaluation choices and how you would compare approaches fairly.

Likely challenges to think through

• Making varied text comparable

  A likely challenge is that combining internal and external datasets may expose differences in format, coverage, or quality. Prepare an approach for checking provenance and consistency before drawing research conclusions; the posting does not specify particular data issues.

• Choosing among model approaches

  Internal models, external APIs, RAG, and post-training offer different implementation and evaluation trade-offs. Prepare to justify a choice against a defined internal task rather than treating a more elaborate approach as automatically better.

• Separating a promising signal from a useful result

  An NLP pattern may support an investment hypothesis without establishing alpha. Prepare to discuss how you would test whether a signal holds up beyond the data used to develop it and communicate uncertainty to research collaborators.

Topics to prepare

• NLP approaches for investment research

  The role applies modern NLP to textual data and formulates research hypotheses aimed at deriving alpha. As preparation, outline a testable hypothesis using a hypothetical set of financial documents, specifying the text signal, outcome, and what result would challenge the hypothesis.

• RAG and agent design

  The posting cites RAG and agents as examples of modern NLP solutions. Compare when retrieval alone, RAG, or an agent would suit a document-based question; identify likely failure points and how you would check the answer against source text.

• Fine-tuning and task-specific post-training

  The posting mentions fine-tuning open-source models and post-training for specific tasks. Review the distinction between prompting and post-training, then sketch a task definition, training examples, and a baseline for a hypothetical text-classification problem.

• LLM benchmarking

  Benchmarking LLM approaches on internal tasks is an explicit responsibility. Draft a comparison plan for two approaches to the same text task, including a consistent test set, success measures, and error review. Those evaluation choices are preparation suggestions, not specified employer requirements.

• Internal models and external APIs

  The role builds generative AI solutions using both internal models and external APIs. Review the trade-offs between these options, then explain how you would compare them for one hypothetical research task without assuming which option Point72 prefers.

Skills to revise

• Explore and combine textual datasets

  The role explicitly involves internal and external text datasets. Practise inspecting two sample datasets, documenting their fields and coverage, and producing a joined, deduplicated set with checks for mismatched records.

• Use Python and SQL for text research

  Python and SQL proficiency is required. Write a SQL query to select and aggregate document records, then use Python to clean the returned text and summarize missing or duplicate entries.

• Prototype a retrieval-based solution

  RAG is cited in the posting. As a preparation exercise, build a small document-question-answer prototype and inspect whether each answer is supported by the retrieved passages; no particular framework is specified.

• Run and compare model approaches

  The role includes post-training and benchmarking. Practise adapting a model to a narrowly defined text task, comparing its outputs with an unadapted baseline on the same examples, and recording representative errors.

• Apply testing and collaborative code practices

  The posting requires general software engineering principles, including GitHub and testing, and a collaborative mindset. Put a small Python text-processing change through a GitHub review workflow with tests for normal and edge cases, and write a concise explanation of the change.

• Prepare an NLP project walkthrough

  Select a project that demonstrates LLM work, ideally RAG, agents, or fine-tuning and deployment of an open-source model. Prepare a concise account of your role, design choices, evaluation, and limitations; these methods are named in the posting.

• Practice combining text datasets

  As a preparation exercise, use public or synthetic text to show how you would explore, join, clean, and track sources with Python and SQL. Do not assume access to Point72’s internal or commercial datasets.

• Draft an investment-research hypothesis

  Write one testable hypothesis about what a text-derived signal might reveal, plus a proposed validation method and possible confounders. The role explicitly calls for hypotheses aimed at deriving alpha.

• Design a GenAI solution

  Sketch a solution for a hypothetical research task using a model, retrieval, or an external API. Explain the trade-offs and how you would handle sensitive data; the posting mentions internal models and external APIs but specifies no required architecture.

• Build a benchmarking plan

  Prepare a comparison of LLM approaches for a specific text task, including a baseline, metrics, test split, and error analysis. Explain when task-specific post-training would be worth testing; benchmarking and post-training are listed responsibilities.

• Gather engineering evidence

  Choose code examples or a repository that show Python, SQL, tests, version control, and data-science workflows. Be ready to explain how you made the work reproducible and collaborated on it.

• Write questions for the team

  Ask which internal tasks need better benchmarks, how text datasets are evaluated for research use, and how NLP researchers collaborate with investment teams. These questions follow from the posting’s research focus without assuming a particular process.