Model training, evaluation & inference

NLP classification workflows

End-to-end classifiers for harmful and biased language, from dataset preparation through application integration.

  • AI training
  • AI inference

The context

Text classification requires more than a model endpoint. The work involved preparing data, experimenting with training, evaluating behavior, and connecting inference to an application.

My contribution

  • Built custom classifiers for harmful and biased language.
  • Owned dataset preparation, preprocessing, model training, experimentation, and evaluation.
  • Integrated inference into application workflows.

Challenges & solutions

Connecting data preparation with a usable classifier.

Owned the workflow from dataset preprocessing through training and evaluation, then integrated inference into the application.

Understanding model behavior before integration.

Included experimentation and evaluation as distinct stages of the workflow, rather than treating training completion as proof of quality.

Pipeline overview

A simplified view of the workflow.

  1. Text datasets
  2. Preprocessing
  3. Training & evaluation
  4. Inference integration

Scope & outcomes

Data, training & evaluation
Full workflow
Application inference
Integrated

Dataset names, specific model architectures, and numerical benchmarks are not available in this public overview. No accuracy figure is claimed.

Building something related?

Tell me about the problem, the constraints, and what success would look like.

Discuss a similar project

Explore more work

Local AI runtime & agent orchestration

Localbots

A Go-based local agent framework and web studio: a manager coordinates specialist bots through one resident Qwen3 model, with no cloud API required after setup.

Event-driven automation & applied AI

Trevy

Event-driven marketing workflows, high-volume messaging, and CRM-grounded LLM content generation.