LLM Trainer

Role Overview:

This position is within a project with one of the foundational LLM companies. The goal is to assist these foundational LLM companies in enhancing their Large Language Models.

One way we help these companies improve their models is by providing them with high-quality proprietary data. This data serves two main purposes: first, as a basis for fine-tuning their models, and second, as an evaluation set to benchmark the performance of their models or competitor models.

For example, for SFT data generation, you might have to put together or be provided a prompt which contains provided code and questions, you will then provide the model responses, and write corresponding Pascal or Delphi code to solve the questions.

For RLHF data generation, you may need to create a prompt yourself or use one provided by the customer, ask the model questions, and evaluate the outputs generated by two versions of the LLM. You'll compare these outputs and provide feedback, which is then used to fine-tune the models. Please note that this role does not involve building or fine-tuning LLMs.

What does day-to-day look like:

- Design, develop, and maintain code modules in Pascal, Delphi, or related dialects.
- Refactor legacy Pascal codebases to enhance performance, maintainability, and readability.
- Create high-quality code-plus-instruction datasets used to fine-tune conversational coding assistants.
- Ensure code samples are syntactically correct, well-commented, and self-contained.
- Write developer-friendly documentation to support model evaluation and human review.
- Evaluate LLM-generated Pascal outputs and provide constructive, structured feedback for model improvement.
- Collaborate with peers on dataset quality reviews and alignment with project guidelines.
- Follow rigorous formatting and quality control standards to ensure data integrity and value.
- Contribute to prompt design, tooling feedback, and optimization of task workflows.

Requirements:

- 4+ years of professional experience in Pascal or Delphi development.
- Strong understanding of procedural programming paradigms, type systems, and BEGIN…END structured blocks.
- Proven debugging, profiling, and performance tuning skills in Pascal applications.
- Solid grasp of Git, version control workflows, CI/CD processes, and testing best practices.
- Excellent written and verbal communication skills in English.

Preferred / Nice-to-Have:

- Experience with FCL (Form Calculation Language) or Intuit's Tax Programming System (TPS).
- Background in TurboTax workflows or other financial/tax software systems.
- Familiarity with domain-specific DSLs or experience modernizing legacy codebases.
- Exposure to AI-assisted development tools, cloud environments (AWS, GCP), or containerization (Docker, Kubernetes)

---