toukan.dev / disciplines / data and ai systems
Data ingestion, machine learning operations, and AI product systems built for production use.
01 / AI product systems
LLM applications, retrieval pipelines, workflow automation, OCR, computer vision, and multimodal product surfaces. We build the production stack around the model: data flows, prompts, structured output, human review, and the software users actually touch.
02 / Data pipelines and MLOps
ETL, ingestion pipelines, dataset versioning, model serving, quantization, batching, autoscaling, monitoring, rollback, and cost controls. The work that turns raw data and model artifacts into production services with clear latency, reliability, and ownership boundaries.
03 / Training and fine-tuning
Supervised fine-tuning, preference tuning, and dataset construction for open-weight models. We define the task, build the training data, run experiments, and measure whether the model actually improves the target workflow.