Scribd
Upload
25 views38 pages

Presentation For AI Engineer

EyeLevel.ai offers a solution to build highly accurate Retrieval-Augmented Generation (RAG) systems in minutes using their GroundX platform, which integrates APIs and no-code tools. The platform aims to reduce common RAG failure rates caused by bad text extraction and missing context, achieving up to 95% accuracy. GroundX utilizes fine-tuned models and semantic object search for effective data extraction and processing from complex documents.

Uploaded by

François Guias
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
25 views38 pages

Presentation For AI Engineer

EyeLevel.ai offers a solution to build highly accurate Retrieval-Augmented Generation (RAG) systems in minutes using their GroundX platform, which integrates APIs and no-code tools. The platform aims to reduce common RAG failure rates caused by bad text extraction and missing context, achieving up to 95% accuracy. GroundX utilizes fine-tuned models and semantic object search for effective data extraction and processing from complex documents.

Uploaded by

François Guias
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 38

THE WORLD’S MOST ACCURATE RAG

WWW.EYELEVEL.AI
BEFORE
I COULD
GROW A
BEARD
GROUNDX
BUILD THE
WORLD’S MOST
ACCURATE RAG
IN MINUTES

APIs + No Code Tools


35%
COMMON RAG
HALLUCINATION
RATE
ON COMPLEX
DOCUMENTS
THREE MOST COMMON CAUSES OF RAG FAILURE

1 2 3

Retrievals Visual formats


Bad text
missing like tables and
extraction
context charts
YOU CAN ACHIEVE

95%
ACCURATE RAG
OR BETTER
BENCHMARK: 98%
1,000 PAGES
OF COMPLEX 64%
DELOITTE
DOCS 43%

EyeLevel.ai LangChain + LlamaIndex +


Pinecone Pinecone
THE POWER
OF SEMANTIC
OBJECTS

100% VECTOR
FREE
Raw Document
GroundX Vision
Model
GROUNDX VISION MODEL EXTRACTS COMPLEX DATA FROM TEXT
DOCUMENTS, SENDS TO PIPELINES TO CREATE LLM READY
DATA

TABLES

FIGURES

DOCUMENT FINE-TUNED
VISION
MODEL
Raw Document
Chunk without
important
context
GroundX
identifies
relevant context
found across a
document
Semantic Objects turn
complex human content
into LLM ready data

Original Text Metadata

Search Optimized Completion


Rewrite Optimized Rewrite
GROUNDX SEARCH AND COMPLETION TEXT
{"type": "Baggage Dimension Diagram", "contents": "This
diagram illustrates the dimensions of an authorized personal
item, specifically a briefcase, with measurements in both
centimeters and inches."}

This diagram illustrates the dimensions of an authorized


personal item, specifically a briefcase, with measurements
in both centimeters and inches.

{"dimension": "height", "value_cm": "30", "value_in": "12"}

The height of the briefcase is 30 centimeters (12 inches).


OCR EXTRACTION
30 cm {"dimension": "width", "value_cm": "40", "value_in": "16"}

( 12 in ) The width of the briefcase is 40 centimeters (16 inches).


40 cm
{"dimension": "depth", "value_cm": "15", "value_in": "6"}
15 cm
The depth of the briefcase is 15 centimeters (6 inches).
( 6 in )
( 16 in )
THE SEARCH PIPELINE

QUERY QUERY WEIGHTED FINE-TUNED


REWRITER SEARCH OF RERANK
SEMANTIC MODEL
OBJECTS
NINE FINE-TUNED
MODELS

9
POWER THE GROUNDX
PIPELINE FOR DATA
EXTRACTION,
PROCESSING, SEARCH
AND RERANK
BUILD PRODUCTION READY RAG IN MINUTES

INGEST SEARCH COMPLETE


TRY EYELEVEL’S GROUNDX
ON YOUR DOCS
https://www.eyelevel.ai/xray
END
THE EYELEVEL PLATFORM

APIs NO CODE

Semantic
Computer Multimodal Query
ETL Object Search Rerank
Vision Rewriter Rewriter
Generator

ADVANCED RAG PIPELINE POWERED BY GROUNDX


BUILD THE
WORLD’S MOST
ACCURATE RAG
IN MINUTES WITH
GROUNDX APIs
FROM EYELEVEL

APIs + No Code Tools


Semantic object search
outperforms vector
similarity
7 FINE-TUNED MODELS

7
POWER OUR RAG
PIPELINE FOR DATA
EXTRACTION,
PROCESSING, SEARCH
AND RERANK
THE EYELEVEL PLATFORM

APIs NO CODE

Semantic
Computer Multimodal Query
ETL Object Search Rerank
Vision Rewriter Rewriter
Generator

ADVANCED RAG PIPELINE POWERED BY GROUNDX


BUILD THE
WORLD’S MOST
ACCURATE RAG
IN MINUTES
MULTIMODAL PIPELINE
Text parser,
summarizer

Document Table rewriter

Multimodal
explainer
Weighted search
Doc sum
Section sum
File keywords
Original text
Visual: how chunking is hard Rewritten search text
Show how information is needed in multiple places
in a document to make sense of it.
THE GROUNDX PIPELINE

1 2 3 4 5

DOCUMENT FINE-TUNED MULTIMODAL DYNAMIC SEMANTIC


VISION TRANSFORM CHUNKER OBJECT
MODEL PIPELINES
OCR HOT MESS
OCR Extraction
EyeLevel Computer Vision DOCUMENT

Extraction
BUILD THE
WORLD’S MOST
ACCURATE RAG
IN MINUTES
DOCUMENT
Source File
GroundX Vision Model
Fine-tuned vision model

You might also like