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.
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).
Job Type: Full-time
Pay: ₦30,579.00 - ₦35,000.00 per hour
Experience:
- PAscal/Delphi: 5 years (Preferred)
- Form Calculation Language: 3 years (Preferred)
- LLM Trainer: 8 years (Preferred)