SAM.AI
ENTERPRISE · AI · AGENCY
SAM.SEHGAL / INDEX
§01 Work
§02 Stack
§03 Practice
§04 About
CET —
● AVAILABLE
§01 WORK
/
02 · AUTORAG
SCALING UP
←
BACK TO WORK
● AI · 02 / 13
AutoRAG
A RAG system that discovers its own optimal configuration.
73%
SCORE GAIN
17
EXPERIMENTS
$15
TOTAL COST
INDUSTRIES
Finance · Healthcare · Research
§ ARCHITECTURE
How it fits together.
CLICK A NODE TO INSPECT →
new config
Benchmark
500 questions
Classifier
Domain routing
Rewriter
Reformulation
Retriever
Vector search
Generator
Multi-model
Hallucination Gate
Validator
Optimizer
7 dimensions
SYSTEM OVERVIEW
Optimizes across 7 dimensions on Meta's CRAG benchmark: 500 questions, 5 domains, 8 question types. 17 experiments for $15 total.
FOR YOU
You want a RAG system that autonomously tunes chunking, embeddings, retrieval, routing, and prompts for your domain.
§ CAPABILITIES
Retrieval optimization
01
Hallucination gating
02
Multi-model routing
03
Benchmark-driven tuning
04
§ IN DETAIL
a.
7 optimizable dimensions including chunking, retrieval, routing, prompts
b.
Baseline 0.208 → 0.360 composite score
c.
Hallucination validator proved more impactful than model upgrades
d.
$0.78 per 100-question evaluation run
← PREVIOUS
AutoAgent
NEXT →
Enterprise SQL Agent