VisionOps Crew: A Multi-Agent Architecture for Computer Vision Operations
VisionOps Crew is a multi-agent computer vision assistant that coordinates specialized agents for planning, dataset curation, model research, ML engineering, notebook prototyping, and implementation support. Built with Google ADK and Antigravity SDK, it turns broad CV requests into structured, executable workflows across Hugging Face, FiftyOne, Jupyter, local hardware, and optional Antigravity SDK execution.
Thursday, July 9, 2026
19 min read
Post-Training Gemma 3 for Earth Observation (EO) Understanding: A JAX Stack + TPU Pipeline for Multi-Label Sentinel Satellite Remote Sensing Scene Classification
This post presents GemmaEarth, a TPU-native JAX post-training and benchmarking pipeline that adapts Gemma 3 4B IT for Earth observation tasks, starting with multi-label Sentinel remote-sensing scene classification using EarthDial and BigEarthNet.
Monday, April 20, 2026
1 min read
From Single-Agent to Production-Ready Multi-Agent Architecture with ADK, MCP, A2A, and AG-UI
This post shows how to evolve a single-agent ADK application into a production-ready multi-agent architecture using MCP for distributed tool execution, A2A for inter-agent coordination, and AG-UI for a unified interactive user experience.
Wednesday, February 18, 2026
1 min read
Building Interactive Agentic Applications Using ADK and AG-UI Protocol
This post explores how to integrate AG-UI with ADK to build usable, event-driven agentic applications, with a practical example that combines weather tools and Google Maps context for real-time interactive experiences.
Friday, December 19, 2025
1 min read
Building a RAG System with PostgreSQL, pgvector, SQLAlchemy, and Gemma/Gemini Embeddings
In this post, we build a Retrieval-Augmented Generation (RAG) pipeline from the ground up. Starting with synthetic data generated by Gemini, we create embeddings with both Gemini and Gemma, store them in PostgreSQL with pgvector, and query them through SQLAlchemy. We also visualize the embedding space interactively to reveal semantic structure. Along the way, we demonstrate how semantic search goes beyond keywords and showcase a multi-agent system that automates dataset generation.
Friday, September 12, 2025
34 min read