The posting
Role Overview
Were hiring a Data Mining Associate to power the backbone of our Legal Data Intelligence systems. In this role, youll work at the intersection of raw data acquisition and ML engineering, transforming messy, unstructured sources into high-quality, consumable datasets optimised for Large Language Models (LLMs).
Youll collaborate closely with ML engineers, responding to data requests and building reliable pipelines that ensure they always have access to clean, structured, and context-rich data for training and evaluation.
This isnt just scraping — its about engineering scalable data mining workflows that fuel next-gen AI.
Key Responsibilities
- Partner with ML engineers to understand data needs and design mining pipelines that deliver training-ready datasets.
- Collect, clean, and normalise unstructured data from legal, regulatory, and public sources into structured formats tailored for LLM workflows.
- Maintain and optimise Python-based data mining systems that handle real-world data complexity.
- Monitor pipeline health, anticipate changes in source structures, and build resilience against disruptions.
- Document data lineage and transformation processes to ensure traceability, reproducibility, and quality control.
- Support annotation and feedback loops, enabling iterative improvements to model performance.
Qualifications
- Have strong Python and data engineering skills with a hackers mindset for extracting insights from difficult sources.
- Understand how to make raw data consumable for ML/LLMs, not just collected.
- Are comfortable with HTML parsing, selectors, APIs, and dynamic content handling.
- Can navigate real-world constraints: proxies, captchas, rate limits, messy edge cases. Bonus:
- Have worked with tools like BeautifulSoup, Scrapy, Selenium, or Playwright for mining.
- Familiarity with LLM data preprocessing (tokenization, deduplication, filtering) and/or cloud data orchestration.
What You Will Achieve in a Year
- Built and owned production-grade data mining pipelines that reliably feed critical legal datasets into LLM training workflows.
- Transformed millions of raw, unstructured documents into clean, structured, and labeled datasets consumed directly by our ML stack.
- Established data quality and lineage standards that allow ML engineers to trace every token in a model back to its source.
- Designed reusable data preprocessing components (deduplication, filtering, normalization) that improve model performance at scale.
- Partnered closely with ML engineers to shape the feedback loop between mined data and model evaluation, helping steer product-level AI behavior.
