SOURCING
Connect to data sources (GCS, BigQuery, APIs, webs).
Robust Over Refined
ArchitectureManifesto
Raggermeister
RAGs BUILT TO LAST.
Latest Entries
The System is the Proof
StrategyManifesto
Beyond Vector Search: Hybrid Retrieval
ArchitectureRetrieval
PRIMARY STACK
VERTEX AI PIPELINES
Orchestration is strategy. Pipelines are used to build robust, repeatable, and maintainable data flows.
Read MoreGEMINI MODELS
The core reasoning engine. Leveraged for its integration, long-context, and multi-modal capabilities.
Read MoreVERTEX AI VECTOR SEARCH
The retrieval core. Built for planet-scale, low-latency similarity search. Robust and battle-tested.
Read MoreBIGQUERY
Where data lives. It is used for structured metadata, analytics, and as a source of truth for retrieval.
Read MoreCLOUD OPERATIONS
If it can't be measured, it's a demo. Cloud Logging & Monitoring are used to prove system reliability.
Read MoreCLOUD RUN
The serving layer. Used to deploy scalable, serverless, and secure RAG API endpoints.
Read MoreCORE LANGUAGES
INGESTION PIPELINE
01
02
PROCESSING
Clean data, extract text from PDFs/HTML, parse JSONs.
03
CHUNKING
Divide documents into semantic fragments (chunks).
04
EMBEDDING
Convert text chunks into vectors using a model (e.g., Gemini).
05
INDEXING
Store vectors (Vertex AI Vector Search) and metadata (BigQuery).