This is a rare prospect to work on LLM/SLM problems that actually ship to production — in a domain where accuracy isn't a nice-to-have, it's the product.
You'll thrive here if you love owning:
→ Fine-tuning pipelines using LoRA / QLoRA / PEFT on domain-specific corpora
→ Synthetic training data generation (JSONL) for healthcare NLP tasks
→ Model evaluation design — precision, recall, and task-specific rubrics for clinical and billing language
→ Quantization and inference optimization (GGUF, AWQ, bitsandbytes) for GPU-constrained deployments
→ Prompt engineering as a systematic, reproducible discipline
→ Benchmarking SLMs against frontier models on narrow, well-defined tasks
Background that maps well: NLP research, LLM fine-tuning at a product company, applied AI in healthtech or regulated domains, M.Tech / PhD with hands-on model training experience.