3. Data, Science & AI

Optimizes BigQuery SQL queries, implements feature engineering, and builds recommendation models using BigQuery ML for efficient data processing and machine learning.

Provides explanations for AI agent architecture, configuration files (e.g., config.json), and components like iterators, reviewers, and prompt systems.

AI-powered GIS assistant for Solano County, California, answering property, zoning, hazard, budget, and regulatory queries through geographic data analysis.

Verifies correct installation and functionality of bazzite-ai, including GPU, CUDA, PyTorch, service health, and network connectivity.

Imports HuggingFace GGUF models into Ollama via hf.co/ prefix, tracks downloads, and enables inference with progress monitoring.

Provides Group Relative Policy Optimization (GRPO) for stable RLHF training, including reward function design and KL divergence constraints.

Enables efficient large language model fine-tuning using LoRA, QLoRA, and Unsloth optimizations for 2x faster training.

Tool for visualizing and curating datasets in machine learning workflows, featuring GPU acceleration and MongoDB persistence via multi-container architecture.

Monitors GPU performance metrics including VRAM usage, loaded models, and tokens per second for Ollama inference.

Comprehensive guide for authenticating with Google Ads API, managing access, and tracking ad performance metrics using Python.

Enables efficient model inference and training via quantization, supporting precision levels (INT8, FP16) and memory optimization.

Comprehensive guide for fine-tuning machine learning models with PyTorch and HuggingFace Trainer, including dataset prep, training, and evaluation.

Provides RAG pipelines for grounding LLM responses using external knowledge, including document chunking, embeddings, and vector search.

Provides a reinforcement learning framework using leave-one-out estimation for stable policy optimization, including reward integration and variance reduction.

Provides a database schema for Google Ads performance data including campaigns, ad groups, and daily/90-day metrics with single-user and multi-tenant support.

Explains transformer architecture fundamentals including self-attention, multi-head attention, and residual connections essential for modern LLMs.

Generates robust GAQL queries for last-90-days ad metrics with automatic field fallback for Google Ads API-rejected fields.

Provides tools for supervised fine-tuning of language models using SFTTrainer and Unsloth, including dataset preparation, chat templates, and optimizations for faster instruction tuning.

Trains reward models for RLHF pipelines, covering preference dataset preparation, sequence classification, and reward scaling for stable reinforcement learning.

Specializes in deploying AI agents using Gemini API, Bedrock AgentCore, and ElevenLabs TTS for AI-powered features in Faiston NEXO.

Enables fast AI model inference with Unsloth and vLLM, covering model loading, output generation, and memory management.

Develops Apache Airflow DAGs for data pipeline orchestration using TaskFlow API, Google Cloud services, and dbt, with dynamic generation and error handling.

Provides OpenAI-compatible API interface for Ollama, enabling seamless migration from OpenAI and compatibility with LangChain and LlamaIndex.

Provides LLM evaluation and prompt optimization using Evidently.ai, including text descriptors, dataset metrics, and LLM-as-a-Judge patterns for classification and generation tasks.