Technology
The technology we build with
Formare AI builds on Google Cloud. The AI layer is Gemini, Google ADK, Vertex AI and MCP; the platform is BigQuery, Cloud Run and GKE Autopilot; applications use React, TypeScript, FastAPI, Node.js and PostgreSQL. This page lists all 64 technologies in 6 groups, each with the write-ups that document it.
Technology
AI & Agents
Models, agent frameworks and protocols behind the agents, voice and search systems.
Google ADK
Multi-agent orchestration: a planner routing to specialist sub-agents for retail merchandising, BigQuery analysis and a personal assistant.
Documented in 5 write-ups
- Senior Merchandiser AI Agent (reference build)
- NL → BigQuery Assistant (reference build)
- LinkedIn Content Agent (reference build)
and 2 more
Vertex AI
Embeddings, Vector Search, the Ranking API and Feature Store for retail search, and Gemini access for content pipelines.
Documented in 10 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- Senior Merchandiser AI Agent (reference build)
- DS Data Catalog (reference build)
and 7 more
Gemini Live API
Real-time voice over WebSocket: browser microphone in, spoken reply and live transcription out.
Documented in 2 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- Gemini Live Audio Assistant (reference build)
MCP
Custom MCP tools and servers, such as reservations through ialoc.ro and LinkedIn actions exposed to any MCP-compatible assistant.
Documented in 3 write-ups
- AI Airport Transfer Platform (client project)
- LinkedIn Content Agent (reference build)
- Personal AI Assistant — Bucharest (reference build)
Gemini 3.5 Flash
The agentic loop with 11 custom tools in the George Travel booking platform, the one client deployment.
Documented in 1 write-up
- AI Airport Transfer Platform (client project)
Gemini 2.5 Flash
Documented in 5 write-ups
- Senior Merchandiser AI Agent (reference build)
- DS Data Catalog (reference build)
- NL → BigQuery Assistant (reference build)
and 2 more
A2A Protocol
Documented in 2 write-ups
Gemini Embeddings
Documented in 2 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- BigQuery Knowledge Graphs (reference build)
Vertex AI Ranking API
Documented in 2 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- Senior Merchandiser AI Agent (reference build)
Vertex AI Feature Store
Documented in 2 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- Senior Merchandiser AI Agent (reference build)
Code Interpreter
Documented in 1 write-up
- NL → BigQuery Assistant (reference build)
Google Search Grounding
Documented in 3 write-ups
- WordPress Deep Research Bot (reference build)
- Personal AI Assistant — Bucharest (reference build)
- Newsletter AI Summarizer (reference build)
Technology
Google Cloud
Where the systems run: compute, warehouse, scheduling, secrets and identity.
BigQuery
The warehouse behind the data catalog, the natural-language SQL assistant and the knowledge-graph build.
Documented in 8 write-ups
- DS Data Catalog (reference build)
- NL → BigQuery Assistant (reference build)
- WordPress Deep Research Bot (reference build)
and 5 more
Cloud Run
Hosting for agents, APIs and web apps, including the George Travel production system.
Documented in 11 write-ups
- AI Airport Transfer Platform (client project)
- Multimodal AI Shopping Assistant (hackathon entry)
- Senior Merchandiser AI Agent (reference build)
and 8 more
GKE Autopilot
Runs the self-hosted BigQuery data catalog.
Documented in 1 write-up
- DS Data Catalog (reference build)
Vertex AI Vector Search
Image and text product search in the shopping assistant. Its sub-300 ms retrieval is a design target, not a measurement.
Documented in 3 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- Senior Merchandiser AI Agent (reference build)
- EU AI Act and GDPR Compliance for RAG Systems on Google Cloud
Cloud Run Jobs
Documented in 1 write-up
- WordPress Deep Research Bot (reference build)
Cloud SQL
Documented in 3 write-ups
- AI Airport Transfer Platform (client project)
- NL → BigQuery Assistant (reference build)
- Deploying Production-Grade Multi-Agent Systems with Google ADK on Vertex AI
GCS
Documented in 1 write-up
- Newsletter AI Summarizer (reference build)
Cloud Scheduler
Documented in 4 write-ups
- DS Data Catalog (reference build)
- WordPress Deep Research Bot (reference build)
- LinkedIn Content Agent (reference build)
and 1 more
Secret Manager
Documented in 1 write-up
- DS Data Catalog (reference build)
Workload Identity
Documented in 1 write-up
- DS Data Catalog (reference build)
Also in the toolkit, with no write-up on this site yet
Technology
Frontend
Interfaces for web apps, dashboards, voice assistants and browser extensions.
React 19
Front ends for the voice assistant, the data catalog and the shopping assistant.
Documented in 4 write-ups
- Multimodal AI Shopping Assistant (hackathon entry)
- DS Data Catalog (reference build)
- Gemini Live Audio Assistant (reference build)
and 1 more
TypeScript
The React front ends, the Chrome extension and the WebMCP extraction agent.
Documented in 6 write-ups
- AI Airport Transfer Platform (client project)
- DS Data Catalog (reference build)
- Gemini Live Audio Assistant (reference build)
and 3 more
Streamlit
Prototype dashboards and assistants: YouTube analytics, a voice prototype and a storytelling app.
Documented in 3 write-ups
- Gemini Live Audio Assistant (reference build)
- MyStoryLand AI Platform (prototype)
- YouTube Analytics Tool (prototype)
Chrome Extensions (MV3)
A context-aware extension that detects what you select and offers Gemini actions, with no background data collection.
Documented in 1 write-up
- Chrome AI Browser Extension (reference build)
Vite 6
Documented in 2 write-ups
- DS Data Catalog (reference build)
- Gemini Live Audio Assistant (reference build)
Panel
Documented in 1 write-up
- MyStoryLand AI Platform (prototype)
WebAudio API
Documented in 1 write-up
- Gemini Live Audio Assistant (reference build)
WebMCP
Documented in 2 write-ups
- WebMCP Data Extraction Agent (reference build)
- Personal AI Assistant — Bucharest (reference build)
Also in the toolkit, with no write-up on this site yet
Technology
Backend & APIs
Services, and the third-party APIs they integrate.
Python 3.12
Agent, pipeline and data services across most builds.
Documented in 8 write-ups
- DS Data Catalog (reference build)
- NL → BigQuery Assistant (reference build)
- Gemini Live Audio Assistant (reference build)
and 5 more
FastAPI
The APIs behind the data catalog, the merchandiser agent and the natural-language BigQuery assistant.
Documented in 5 write-ups
- Senior Merchandiser AI Agent (reference build)
- DS Data Catalog (reference build)
- NL → BigQuery Assistant (reference build)
and 2 more
Node.js
The George Travel backend and the WebMCP data-extraction agent.
Documented in 2 write-ups
- AI Airport Transfer Platform (client project)
- WebMCP Data Extraction Agent (reference build)
Stripe
Payments in the George Travel booking platform.
Documented in 1 write-up
- AI Airport Transfer Platform (client project)
Express 4
Documented in 1 write-up
- AI Airport Transfer Platform (client project)
Google Maps Platform
Documented in 1 write-up
- AI Airport Transfer Platform (client project)
Google Calendar API
Documented in 1 write-up
- AI Airport Transfer Platform (client project)
Gmail SMTP
Documented in 2 write-ups
- AI Airport Transfer Platform (client project)
- Newsletter AI Summarizer (reference build)
AeroDataBox
Documented in 1 write-up
- AI Airport Transfer Platform (client project)
YouTube Data API
Documented in 1 write-up
- YouTube Analytics Tool (prototype)
LinkedIn API
Documented in 1 write-up
- LinkedIn Content Agent (reference build)
WordPress REST API
Documented in 1 write-up
- WordPress Deep Research Bot (reference build)
Technology
Data & Databases
Stores, modelling and the analysis libraries.
PostgreSQL 16
Booking data, with advisory locks to prevent double-bookings, and metadata for the data catalog.
Documented in 3 write-ups
- AI Airport Transfer Platform (client project)
- DS Data Catalog (reference build)
- Deploying Production-Grade Multi-Agent Systems with Google ADK on Vertex AI
Neo4j
A knowledge graph built from BigQuery product tables and queried through a natural-language-to-Cypher agent.
Documented in 1 write-up
- BigQuery Knowledge Graphs (reference build)
Pandas
The analysis agent in the natural-language BigQuery assistant, and the dashboards.
Documented in 2 write-ups
- NL → BigQuery Assistant (reference build)
- YouTube Analytics Tool (prototype)
SQLAlchemy
The data layer of the data catalog's FastAPI backend, with Alembic migrations.
Documented in 1 write-up
- DS Data Catalog (reference build)
SQLite
Documented in 1 write-up
- NL → BigQuery Assistant (reference build)
Alembic
Documented in 1 write-up
- DS Data Catalog (reference build)
NumPy
Documented in 1 write-up
- BigQuery Knowledge Graphs (reference build)
Plotly
Documented in 1 write-up
- YouTube Analytics Tool (prototype)
LightGBM
Documented in 1 write-up
Also in the toolkit, with no write-up on this site yet
Technology
DevOps & Tooling
Packaging, deployment, testing and browser automation.
Docker
Container images deployed to Cloud Run, as in George Travel and MyStoryLand.
Documented in 3 write-ups
- AI Airport Transfer Platform (client project)
- Newsletter AI Summarizer (reference build)
- MyStoryLand AI Platform (prototype)
gcloud CLI
Service accounts and IAM, Vertex AI endpoints and Cloud Run deploys, scripted rather than clicked.
Documented in 4 write-ups
- DS Data Catalog (reference build)
- Newsletter AI Summarizer (reference build)
- Deploying Production-Grade Multi-Agent Systems with Google ADK on Vertex AI
and 1 more
Cloud Build
CI/CD for the data catalog: automated builds and deploys.
Documented in 1 write-up
- DS Data Catalog (reference build)
Chrome DevTools Protocol
Lets the WebMCP agent discover and call a page's tools.
Documented in 1 write-up
- WebMCP Data Extraction Agent (reference build)
Also in the toolkit, with no write-up on this site yet
Read next
See the technology in use
Implementation articles
How each build was made: architecture, code and the trade-offs, with the numbers labelled as measured, targets or vendor-reported.
Read the articles
Services and pricing
What we build with this stack, how an engagement runs, and what it costs.
See the services