Please wait while we prepare your content
The Pramana Framework
RAG is the practice of preventing them.
The Pramana Framework is a four-book curriculum that takes you from your first embedding to a production-grade multi-tenant RAG system — the architecture that stops language models from answering questions they should not answer.
No signup required · Available on Kindle Unlimited
The Problem
A language model that has no retrieved context does not say “I don't know.” It answers — fluently, confidently, incorrectly. No error is thrown. No alert fires. The HTTP status is 200.
This is Confident Drift. It is not a prompt problem. It is an architecture problem. The Pramana Framework is the architectural answer.
The model answers wrong. The HTTP status is 200. No alert fires. Your monitoring shows green.
"Only answer from the documents" is an instruction. An instruction the model can ignore. Architecture cannot be ignored.
The retriever decides what the model is permitted to speak from. That decision is where drift is stopped.
The Curriculum
Each book ships a working system. Each chapter adds one component. By Book 4 you have built a multi-tenant, GDPR-compliant enterprise RAG platform.
Build Your First AI App
Build ShopBot — a production RAG system for an Indian fashion store — from a single Python file to a live Railway endpoint. Embeddings, ChromaDB, retrieval, prompt engineering, RAGAS evaluation.
Faithfulness 0.71 · Context Precision 0.82
Get on Amazon →Real Apps, Real Data
Hybrid BM25 + dense retrieval. CrossEncoder re-ranking. Redis conversation memory. RAGAS evaluation harness. Migrate from ChromaDB to Qdrant without breaking production.
Architectures & Optimization
HyDE for vague queries. CRAG for self-correcting retrieval. LangGraph for adaptive routing. Fine-tuned ONNX-quantised model trained on your catalog.
Enterprise Systems
Six tenants. Isolated Qdrant collections. AWS KMS encryption. GDPR/DPDP deletion. DynamoDB immutable audit log. Redis rate limiter. Locust load testing.
Available on Kindle Unlimited
Free Preview
Arjun had been proud of ShopBot. It ran without errors. It answered questions. It had never crashed in six weeks of testing.
He found out what it had been doing instead.
“Every hallucination is a grounding failure. RAG is the practice of preventing them.”
What You Will Build in Book 1
The Author
Founder, zUdyog · RAG Practitioner
Balaji Hambeere built the Pramana Framework while engineering production RAG systems for zUdyog, his Indian fashion retail platform.
The problems in these books — retrieval failures, Confident Drift, multi-tenant isolation — are real problems from a real system. ShopBot is not a fictional example. It is the architecture, described in teaching language so any engineer can build it.
The word Pramana comes from Sanskrit: valid source of knowledge. That is what a RAG system is, when it is built correctly.
New chapter releases, code updates, and architecture notes — delivered before the book publishes.
No spam. One email per chapter. Unsubscribe any time.
Start with Book 1. Build ShopBot from scratch. Understand the Grounding Layer.
