AI & Machine Learning Technologies
From frontier hosted models to CPU-only local inference and fine-tuning — see AI & Machine Learning Engineering for the full service breakdown and real-world results.
Frameworks & LLM APIs
PyTorch & TensorFlow
Deep learning frameworks for real model fine-tuning, evaluation, and inference work across both hosted and self-hosted models.
HuggingFace
Model hub and tooling for sourcing, evaluating, and fine-tuning open-weight models before they go into a deployment pipeline.
OpenAI, Anthropic Claude & Google Gemini SDKs
Direct API integration of frontier hosted models into business workflows and applications, including the LLM-classification stage of SPAM Escudo’s three-phase detection pipeline.
Vector Search, Local Inference & Applied ML
PGVector
Embedding-similarity search inside PostgreSQL, powering SPAM Escudo’s spam-detection pipeline without standing up a separate vector database.
Local Model Deployment & Fine-Tuning
Evaluating quantized open-weight models (Qwen 3, Phi 3.5) for an in-progress invoice-detection pipeline that extracts data from emails and WhatsApp messages, and fine-tuning a 270M-parameter Gemma 3 model for a classification task — all running on CPU, no GPU required.
Applied ML & Data Analysis
scikit-learn, pandas, and numpy for classical ML work: TF-IDF and text clustering that shaped CLEAN_Update’s roadmap, and model-weight/probability-calibration tuning that improved Cerberus ID’s fraud-detection accuracy.
Every capability above is backed by real client and product work — see the AI & Machine Learning Engineering page for the full picture, or get in touch to talk through your project.