AI Trainer — LLM Training, Evaluation & Data Pipelines
OtherImageRLHF
At NileCare Solutions (May–Nov 2025), I designed instruction-tuning datasets from scratch — defining task taxonomies, writing labeling guidelines, and building QA pipelines (schema validation, deduplication, leakage detection) to reduce noisy samples. I developed LLM behavioral evaluation suites using curated test sets and automated scoring scripts covering helpfulness, safety, factuality, and grounding. I also built TypeScript/Node.js annotation utilities to scale the labeling and review process, and ran reproducible training experiments using Hugging Face Transformers + PEFT/LoRA packaged with Docker.
At NileCare Solutions (May–Nov 2025), I designed instruction-tuning datasets from scratch — defining task taxonomies, writing labeling guidelines, and building QA pipelines (schema validation, deduplication, leakage detection) to reduce noisy samples. I developed LLM behavioral evaluation suites using curated test sets and automated scoring scripts covering helpfulness, safety, factuality, and grounding. I also built TypeScript/Node.js annotation utilities to scale the labeling and review process, and ran reproducible training experiments using Hugging Face Transformers + PEFT/LoRA packaged with Docker.
2025 - 2025