Slides: Structuring the Unstructured - Cedric Clyburn, Red Hat
Source Video
Structuring the Unstructured - Cedric Clyburn, Red Hat
Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.
Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.
Extracted Slides
OCR text:
~~ a Al engincer
icone | World's Fair |
Ais} [o) of =) g
Structuring the Unstructured: Advanced
Document Parsing for Al Workflows
AT Engineer: 2026
(OX-to fool an mee
Sen:or Developer Advocate ;
Bacenae sesame sink ©| _ rs
OCR text:
We've got a lot to cover today!
pe ©
a
Wait, so 85% of the
world’s data is...
unstructured?!
OCR text:
We've got a lot to cover today!
, Azure Al Document Intelligence
A . > > 8) aomnmeemeeene einen
PD r UamaPerse: Transtorm unstucurea «= AMazon Textract
F ) But current solutions are — S#att LM ovemizedfomats treme eros temtet te manera tna meneens
proprietary, and require
a sending your private data!
Wait, so 85% of the a oS ee se a
world’s data is... "een so seen me re
unstructured?! let's learn about l : nit Nowe
extraction, parsing, ek or » er psis
chunking, and much more! ) ae “gD TE, fa
( tee Ee
Se a
So, how can we easily parse
bes graphs, tables, etc to formats
OCR text:
Data is the key ingredient behind Al applications!
Chaptar 2 Creating an Amazon $3 cheat ind [=r eeeesee
steno S Shae
— vem ane Financial
“Tote — =. _ Documents
— Technical Meeting Minutes
Documentation e
OCR text:
Data is the key ingredient behind Al applications!
a
ea — iEeses-+ Powering:
a - ceang anamaens3eerie9 : Sse RAG (Document Q&A)
notebook coms tae
. St Fine-Tuning
cecpnenuene tenes Hoegmte ge nah tener eT OO aT Financial
ee ae Documents 7 ete.
oe —— — SEs
_ cee Meeting Minutes 15s Wcenbee pitin ibe
nore wmen we eeeteneinns ~ amie ey J. Hugging Face
Do, a nVIDIA. .
ake geen . &3 databricks
are MMeta -s .
Technical ere nn .
Documentation / H Mi
+ much more! es Go gle
Knowledge Base oe
nice his
OCR text:
Data processing & prep is quite important!
& gqurovduptal ©
© lol over 20 screnbtic papers now feature the
nontensical term ‘vegelatire electron
frcroscopy”
all because an Al mebnierpreted a 1959 article,
Merging ‘vegetalne and electron mcroscopy’
” - . ae . - 4
0 tet Fe ae Gan me a
bottom Ih fy mr epaewm ead enastnet the
to Ue overlies rerine maorengy Noo
pore Bipriaed In Cle cesmpoream was edtened
— ed} west part nf Th wee med known whey
ted ew antes asta os eee ce aneeber rest
Gee Cpa en oY
OCR text:
Data processing & prep is quite important!
WETU UOTUUMARD IU) BL CALIECS FD BUTE UN RCSCE UT LY pe. 16 Bas CUUCIUUOU ORS BL EOE
integrated at pH 7.0. Peptide was released part of the sporangia) wall was dissolved away
which eetabliahed that the coats contained sub- to allow release of the spore. It appears likely
strate for the lytic enzyme present in spores. that the exosporium of B. cerews does not hare
Peptide was also released [rom spore costs of B. a composition samilar to that of the vegetative
megatervum by the action of the enzyme from B. cell wall, from the results obtained by Dr. J. R.
‘The lytio entyme did not how fe tee . .
chanel bog The spore develops in the vegetative ced, whach thus becomes a sporangum It a by ho mans certain what happens to the
The spore develops in the vegetative cell, snaetatne cell wall when the spore is released In Clostncum species tf appears that at least part of thes structure «6 retained as an
which thus becomes a sporangium. It is by no Gutter Memb ane ound the epore It the opine of some workers that the wal of the sporuaating ced forms the expeporium wher
a ae Pai . G0SIS a5 an Outer Coat around spores of several Bac dus species Spores of several vaneves of B cereus had exosporia whereas these
cel) wall when the spore is released. In Clee. 9 S7UCturet appeared to be advent from spores of B megotenum and B suptin It seers, however, that in Bacwus speces at least
fridium species it appears that at least pert of (it seater pert of the vegetative cet watts dssolved away before the developed spore 6 released If thes 6 true, then soluble
this struct is retained as an outer membrane Components contamng the characterst< cometuents shoukd appear in the mechan during wore release Cutture filtrates from &
4 the It is the . of some Cereus organs at various stages of growth and sporulabon were hydrolyzed and the hycrolyzates anatyred for amino sugars and
» % chamingpuneld acid (28) Results showed that a large increase n the concentration of these substances in the cutture filtrate
workers that the wall of the sporulating call ood dung ipore release (lable 2), they were found to be present in a nonckatyzable peptide of the charactecnc type It was
forms the exosporium which existe a0 an outer conciaded that at least part of the sporangaal wal was dasolved away fo ahow release of the spore It appea’s baely that the
examporum of B cereus Goes not have a compoution smiw to that of the vegetative cel wast, from the cesults oblaned by Dr JR
Norns of Leeds Uneverstty (personal communcation) He treated spores with a mgity ace preparation of lyt enzyme fr
Cereus Spores and examined the effect by means of glection mcroscopy ~~ @
OCR text:
So, let's try a simple PDF parser...
S J : : 5 LE 2 : 8 : S ra Y Very fast and cheap
eo ap 258 ee oF co X Incomplete
= a Soyieie: i X Loss of structure
| L alco re ~ Unfit for most use
a ove ae Sue cases
OCR text:
But powerful frontier models? Not bad!
ed atid ete eum
= SEED BES on Ce Y Good quality and
Se gba ge be be BO robustness
= ns SS se som ! Expensive (for now)
wee SEE we 2 FE. 1 Hard to achieve consistent
= SUSIE ESS - Cee structured output
Il Ee - mes i ew ieee a
Sites Ses soe se nee ~ Possible hallucinations
i worrioseneneaan Very costly at scal
WEES AUST EEEEEE See a” not always faithf
OCR text:
Maybe there's a middle ground... Welcome to Docling!
Renee: ae SES == YY Good quality
Ho Pilectissis Cay mn te orielR is leet i= Fast and cheap
e- nee hg mmm foe ee fuse 2 Efe go | Fully local operation
— aeeRiewtome aim fe mfp ig tie pizis Structured format output
me EU 22 ~e Cost-effective at scale, wit
(Ee » sm few anjonle se eet 8 Cost-eff le, with
ame see ees me ee DB ee BODE consistent representation
I Soe ele ssinels BOF ee Be and high quality
Sieacmeios T eeeey —= -
repute ered ds AML foo ry ad aloo be Ene
OCR text:
Docling: Get your documents ready for gen Al
An open source processor using advanced vision models * OCR
Parsing of multipie document formats incl. PDF, iC ™
DOCX, XLSX, HTML, images, and more Oi ced thd
mee Gen
» Advanced PDF understanding with page layout, j= a - | | r
| '
reading order, table structure, code, formulas, - ast
image classification, etc
» Plug-and-play ecosystem integrations
» Local execution for sensitive data and
air-gapped environments
OCR text:
Docling: Scale, cost, and performance
5
aan
me >
ie iiss ecg en Amram Mee & 475,019,140 PDFs parsed end-to-end
— 2 ree (Ss Aas CRRA OR PTERRS, Q 1,733 languages represented
[O mee tee weet Beene enews ror nes iJ ~3 trillion tokens (~2,918B) extracted
enclmis =—— SE esac M1 3.65 T8 of high-quatity, deduplicated text
. ~ ~ : ie DR TE Ve “* Data spanning 2013-2025 across 105
Aesdl fens Beansece wiles CommonCrawl snapshots
:
e
(=) (S) Fine PDFs 918/368*750/35 = 50
: a eRge ERORS
= “ i
seeeeeetee mee | ee . Docling is 50 times more
oe — — cost-effective than VLMs!!
6 Aen snes mary pe ae - k&
O mame tions to
Fine PDFs rasiors
ce
OCR text:
. < ‘
Docling: More than simple Document Conversion
4
—_-_ = = we
z o es Ss o
’ . . & .
e
i + me
The proveieg chart enage m & bay chart Pat Ongiove dete acrrs toe quarters O1.'$ OF 15 O)15 end 0413 The chert ome
e= -_ Aol Mave 8 spaced soe The Fan represenes he Quarters she hie yams Wows 2 percentage teats Fanomg ho 7 10
oa baal wan. TOON Cart bee 0 segrerted ren GMeren cours each ~wpresenting 2 wpectic caegary Agency POMMES (ight bis, Non Agency
i Fe gene ce pp emmamium manner seman eitoaw ee FOMES Bart Bie) MSA (yetoe) Mangage Leen Condut ined: and Comers OrEnge: ING? 1S the Agency RARBS canegory
= el ee eee ee oe 2a Pe large comoreng 65° of he telat totowed by Non Agency RS of 4 MEA of 127. Uorigege Loan Conant of 11%
| Sea ae aNd Commercat at 1% Moving © 02'S Pa Now Agency RMB calgary mumanet 29% becomrg Pe lenge sepnert
STI SS white Agency RMBS decreases 12 44% USM and Ulerigage Loan Condt BoP naman at 11 and Commercial tan at 1S
seAEA ENRON ARON 0315 Re NomAgency RMBS category Later necresees 10 0% ofl Agency RUBS decreases 19 415 USK reas 10 14
| Mongage Loan Comma neveated © TIN and Commerctl 1259010 4% By O4-7S te Non Agency RMBS calmgary a te
ass 2 leryeel af 770 eh Agency POMES af 2. MIA and Moryage Loan Cone bah decrease 14% and 18% papacy ofte:
il i i i ‘Commmertul rcreaees 10.0% The chart veuaty epreeern Pe Changes 1 Pe preportone of Pose categoren Der Pe tose
gare The mamrase sete ofserved 6 45% toy Agency MMS 9 113 and Pe mewrerh coke 8 1h to Comvnerces 7 bor
vos 7 G1-15 ang. G2 13 The range tor wack caimgory ae a6 Kok Agency RUBS ranges Fen 31° 43 Hon Agency MBS
was Wem 271 AIS MEA her TON 2 TEN Mongage Lean Candua form 119. 1 “@1 and Commenced hee TN me The cnet
Stee a ee (eect BinHe Sneing properties Of Reus Calagaren ove fe speced wre parc _
= ar cart a
se Se ee BEE ee reecs
a a Be | So SS ee See Cure Agency, —NanAgemry «MR Merigege Lenn Conta
ra paca ean ow ow 7 wom
| Saal, mon
Soe Se eae ae eee as ae ™ wa mn c
popenacem cheiennrr ma _ oom nm nw wm
OCR text:
Docling:MorethansimpleDocumentConversion
hello
1010A**01+
'billnumber':'01234',
'total invoice price':550,
'currency of total invoice price':'usD',
'name ofinvoice addressee':'Jonathan Patterson'
'name of invoice sender':'Eventure Event Planner
invoice_dict=
:Jaqunu13q
"string"
"totalinvoice price":"float",
"currency of total tnvoice price":“string”
"name of invoice addressee':"string”,
OCR text:
.
e ) 1 ibe-grarvie-community + docling-workshop = 26 ~ GQ type + te veoh a- 7. On 2 e
<> Code ©) ttsues SL Pultrequests Agents ©) Acbom = #3 Projects «Security and quality 9 oY Insights
docling-workshop °>: @Q wach to - Yroa oe - Stwrea 27+
P mein ~ D PF iBeareh Cy tag Q Geteme + Addtte - Ge Source code for Dochng Workshop.
OC aen-grante-community github 10f
@] dthorgrave Mere cu segues! HID Hoe -tre-grante Comers ty Ueberdabet MB te 2 tres, D106 Comments
wobihop gante — decing 8
es) Bop gt uhrodea! actor tM git actions g¢tup ar Bays age TD aestne
Be docs ASDC CHID COCR ts Lote EE SHOE Da Ore woe ante AD Apache BO cern
G Coste ot cordunt
ME notebooks Serco FAS al gteuidey a th Cos cap bt be cars, attawes
BU Coatetegt ony
B& scrptsregenerale_fntures. Tiaitch Rep wale Tede to gante 67 ab 2 acees aie N Atel
Fi Suse reegentes
misc Hierrcne ogy ots Lrmartts age x SARS Cgaer te
TT 2 vace
CO gitgnere Here egg. ote Leon hs ager © Vwatireey
CO mackdowntet-cl2 yar Rectone netic utp! ocr ae: Lrorth sage YoOtes
Re Vyear ona
D poutne ym STS ABM sot batenp ath ene are poutre Droetr ss ayer
Heer Drepostoty
{}) pre-commut-contig yaré Upiiates to deperdane! for yrcuped ocdstes beays 990
Deployments 99
D) seelicheck-en- custom txt Wee Chosen BAG a4 cab a eto montce Goes ard Ce Devoe ths age
@ oihud-pages +,
CO speticheck yami ait A anton bat ag adh cachet are pont oe death sage
0 VCENSE ata cured Convibutors 4
() README ma Satih Rep cate mage toguete 47 st 2 wees age 3& bihergrave &. eargtare
CY constramts tet Sitch te Cnramaud ‘er arcterstere brocths aqe z mingxrhao tngi uae Shae
Oy niaees wr! tho fe one eee WEE ae ak Dd ptr, pare eng oce are) © cy peaae ths ae, FR re
OCR text:
Q
B
Conversion.ipynbMX
Token_Cost_Comparison.ipynb M
Chunkless_RAG.ipynb M
Chunking.ipynb
DOCLING-WORKSHOP
notebooks
fixtures
Add Code Add Markdown|Run All Restart Clear All Outputs|Jupyter Variables Outline
venv (3.13.7) (Python 3.13.7)
ndno<
ImportEssentialComponents
gitignore
Chunking.ipynb
Chunkless_RAG.ipynb
from pathlib import Path
Conversion_Colab.ipynb
Conversion.ipynb
M
Core Docling imports
MCP_Agents.ipynb
M
from docling.document_converter import DocumentConverter
Ollama RAG.ipynb
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions
RAGipynb
from docling.document_converterimportPdfFormatoption
Serving.ipynb
Token_Cost_Compari.M
#For advanced features
scripts
src
#For data processing and visualization
gitignore
import matplotlib.pyplot as plt
!markdownlint-cli2.yaml
!.poutine.yml
Create output directory
!pre-commit-config.yaml
output_dir=Path("output")
output_dir.mkdir(exist_ok=True)
F.spellcheck-en-custom.txt
√0.0s
Python
!.spellcheck.yaml
Fconstraints.txt
LICENSE
BasicDocumentConversion
!mkdocs.yml
pyproject.toml
MinimalExample
①README.md
The simplestway toconvertadocument:
OUTLINE
OCR text:
.
Par SRL LBs # Bs oe a ‘y rey
Your turn: Try your own document
tare Lege wey foe gee fey ee ees rt a
oe ee re an eee ee a et
7 Table 1 Versons and cuafiguraion option comidered for cach bested anset. * de
Lae
Awat Veron OCR Layout Tables
, Dochng 242 bay OCR defuk Tabbe burp
Maer 0410 0 Suna” Sefauk deft
|| Mind 094 mao” Soclayuse yobo raged table
Uerinatered 0165 fa res wah Ube arwctere
7 Documents per Category © When running expen
bene reees CUDA axcelention
eee vane 3 Marcas easure the GPU is vi
9, snfercexe are instal!
a Law ord Megs ine
a Table 1 pronides aa 0
wots Laon opbons we con
SA Results
s * sot F CONP ae .
wy
Dages per Category | oy
fe trey ear :
rom RT Qe “y
/ Lim One Royse ney , .
i:
wpe
i Poo] I eo
raat Rapacty CORR ree " *
: a is
7
Catone es 7
Reread Ranger me Peterty tee
me CCAR wl me Swe ae
Lae ma Regan ore vets
om Hey Higure 3. Distnbuon of
ordered by number of
OCR text:
Q。
Conversion.ipynbMX
Token_Cost_Comparison.ipynb M
Chunkless_RAG.ipynb M
Chunking.ipynb
DOCLING-WORKSHOP
Al-Ready Data with DoclingM Basic Document Conversion>M Document Structure Exploration>#Document metadata-important for tracking
notebooks
fixtures
AddCodeAddMarkdownRunAll Restart ClearAllOutputs|JupyterVariablesOutline
venv (3.13.7) (Python 3.13.7)
ndno<
SectionHeaderItem:Abstract
gitignore
TextItem:We introduce Docling,an easy-to-use,self-contained,MITlicensed,open-source toolkit for document conversion,tr
Chunking.ipynb
TextItem:Repository-https://github.com/DS4SD/docling
Chunkless_RAG.ipynbM
SectionHeaderItem:1Introduction
TextItem:Converting documentsback into a unified machineprocessable format has been a major challenge for decades due to th
Conversion_Colab.ipynb
TextItem:*These authors contributed equally.
Converson.pynb
M
MCP_Agents.ipynb
M
Ollama RAG.ipynb
ExportFormatsandOptions
RAG.ipynb
Serving.ipynb
Doclingsupportsmultipleexportformatswithvariousoptions:
Token_Cost_Compar.M
Add Code
Add Markdown
scripts
src
#Export to different formats (various options available,but called with default ones)
gitignore
markdown_text=doc.export_to_markdown()
!.markdownlint-cli2.yaml
html_text=doc.export_to_html()
!.poutine.yml
json_dict=doc.export_to_dict（)
doc_tags=doc.export_to_doctags()
!.pre-commit-config.yaml
F.spellcheck-en-custom.txt
#Save different formats(various options available,some showm)
!.spellcheck.yaml
doc.save_as_markdown(
constraints.txt
output_dir/"document.nd",
LICENSE
image_mode=ImageRefMode.PLACEHOLDER,
image_placeholder="<!--my image placeholder
!mkdocs.yml
pyproject.toml
①README.md
OUTLINE
JSON
OCR text:
Q
Conversion.ipynbMx
Token_Cost_Comparison.ipynbM
Chunkless_RAG.ipynb M
Chunking.ipynb
DOCLING-WORKSHOP
ntsinto Al-Ready Data with Docling>M Working with Tables>M Basic Table Export >print(f\nDocument contains{len(table_doc.tables)} tables)
notebooks
fixtures
Add CodeAdd MarkdownRun All Restart Clear All Outputs|Jupyter Variables Outline
.venv (3.13.7)(Python 3.13.7)
output
#Save as HTHL
gitignore
withopen(output_dir/f"table_(table_idx).html",
"）asfp:
Chunking.ipynb
fp.write(table.export_to_html(doc=table_doc))
BChunkless_RAG.ipynb
[es]
√0.0s
Python
Conversion_Colab.ipynb
Usage ofTableItem.export_to_dataframe()withoutdoc'argument isdeprecated.
Conversion.ipynb
M
MCP_Agents.ipynb
M
Document contains 8 tables
Ollama RAG.ipynb
RAG.ipynb
##Table
Shape:(4,4)
Serving.ipynb
Token_Cost_Compari..M
Cauldron
LLaVa-OneVision
Cambrian-7m
scripts
General
276.5K
881.3K
1.8M
src
Language/Captioning
202.1K
N/A
NA
gitignore
Math/Science/Reasoning
178.4K
318.0K
354.5K
!.markdownlint-cli2.yaml
Image Comparison
188.9K
N/A
NA
!poutine.yml
!.pre-commit-config-yaml
F.spellcheck-en-custom.txt
##Table1
!.spellcheck.yaml
Shape:(4,4)
Fconstraints.txt
LICENSE
Cauldron
LLaVa-OneVision
Cambrian-7m
!mkdocs.yml
General
812.7K
2.0M
7.9M
pyproject.toml
Language/Captioning
203.3K
1.2M
1.8M
①README.md
Math/Science/Reasoning
765.1K
464.8K
802.0K
Image Comparison
237.9K
N/A
N/A
OUTLINE
OCR text:
Q
B
Conversion.ipynbMx
Token_Cost_Comparison.ipynb M
Chunkless_RAG.ipynb M
Chunking.ipynb
DOCLING-WORKSHOP
notebooks
fixtures
AddCodeAdd MarkdownRun All Restart ClearAll Outputs|JupyterVariablesOutline
venv (3.13.7) (Python 3.13.7)
output
gitignore
Chunking.ipynb
InspectingPicture Content
Chunkless_RAG.ipynbM
Conversion_Colab.ipynb
Conversion.ipynb
M
D日
MCP_Agents.ipynb
M
definspect_pictures_with_inages（doc:DoclingDocument,image_size=(6,4)):
Ollama RAG.ipynb
pisplay pictures inline with their text content.
RAG.ipynb
foridx,pictureinenumerateiPictureItem](doc.pictures):
Serving.ipynb
print(f"\n('=′60)")
Token_Cost_Compari..M
print（f"Picture(idx)")
print（f"('='*60）")
scripts
）src
#Display the image
gitignore
try:
!.markdownlint-cli2.yaml
img=picture.get_image(doc)
ifimg:
!poutine.yml
plt.figure(figsize=image_size)
!.pre-commit-config.yaml
plt.inshow（img)
F.spellcheck-en-custom.txt
plt.axis('off')
!.spellcheck.yaml
plt.title(f"Picture(idx}")
Fconstraints.txt
plt.show()
LICENSE
print（f"Could not display image:(e)")
!mkdocs.yml
pyproject.toml
#Display metadata
README.md
caption=picture.caption_text(doc)
ifcaption:
print（f"\nCaption:(caption}")
OUTLINE
OCR text:
Visualizing Document Layout with Bounding Boxes
eee eee Sa ae ra re eNO et TED bet ee ee ee ea ae
(howe oh fe Mog Bog gee bY oe ue 4° cc ee eee
DwaiHpy An Efficient Open-Source Toolkit for Al-driven Document Conversion
Middiaos Livathinos ©, Christoph Auer ©, Mauksym Lysak, Ahmed Nassar, Michele Dalfi,
Panagiotis Vagenas, Cesar Berrospi, Matteo Omenetti, Kasper Dinkla, Yusik Kim,
OCR text:
Q。
B
Conversion.ipynbMx
Token_Cost_Comparison.ipynbM
Chunkless_RAG.ipynbM
Chunking.ipynb
DOCLING-WORKSHOP
3nced Features: Enrichment>MYour turn:Prompt the vision model >from docling.datamodel.pipeline_options import PictureDescriptionApiOptions
notebooks
fixtures
AddCodeAdd MarkdownRun All RestartClearAll Outputs|JupyterVariablesOutline
venv (3.13.7) (Python 3.13.7)
output
WecanalsorunitusinganOpenAl-compatibleAPi likeOllama.
gitignore
Chunking.ipynb
D
Chunkless_RAG.ipynb
from docling.datamodel.pipeline_optionsimportPictureDescriptionApioptions
Conversion_Colab.ipynb
Conversion.ipynb
ifRUN_LOCAL_OLLAMA:
M
MCP_Agents.ipynb
#Configure enrichment pipeline
M
enrichment_options=PdfPipelineOptions(
OllamaRAG.ipynb
do_picture_description=True,
RAG.ipynb
enable_remote_services=True,
Serving.ipynb
picture_description_options=PictureDescriptionApioptions(
Token_Cost_Compari..M
url="http://localhost:11434/v1/chatcompletions",
params=(
scripts
"model":"granite3.2-vision:2b",
src
"max_completion_tokens":200,
gitignore
!.markdownlint-cli2.yaml
prompt="Give a detailed description of what is depicted in the image"
timeout=60,
!.poutine.yml
!.pre-commit-config.yaml
generate_picture_images=True,
F.spellcheck-en-custom.txt
images_scale=1.0,
!.spellcheck.yaml
Fconstraints.txt
converter_enriched =DocumentConverter(
LICENSE
format_options={
!mkdocs.yml
InputFornat.PDF:PdfFormatoption(pipeline_options=enrichment_options)
pyproject.toml
①README.md
enr_result=converter_enriched.convert(docling_paper)
OUTLINE
enr_doc=enr_result.document
OCR text:
BConversion.ipynb M
Token_Cost_Comparison.ipynbM
Chunkless_RAG.ipynb M
Chunking.ipynb
DOCLING-WO.
kless RAG actually earns its keep> query_2 = What was Red Hat's revenue growth in 2025 and how did it contribute to IBM's overall software segment?
notebooks
fixtures
Add CodeAddMarkdown|Run AllRestart ClearAll Outputs GoToJupyterVariablesOutline
.venv (3.13.7)(Python 3.13.7)
output
HowchunklessRAGworks
gitignore
Chunking.ipynb
Chunklessretrieval isa four-step loop:
Chunkless_RAG.ipynbM
Conversion_Colab.ipynb
Conversion.ipynb
M
Document outline
MCP_Agents.ipynb
(per-section summaries)
M
Ollama RAG.ipynb
RAG.ipynb
Serving.ipynb
Token_Cost_Compari.M
1.SELECT
-LLM picks the most relevant unvisited
scripts
section by reading the outline + query.
src
markdownlint-cli2.yaml
!.poutine.yml
2.FETCH
-Pullthe full textof thatsection's
!.pre-commit-config.yaml
subtree from the DoclingDocument.
F.spellcheck-en-custom.txt
!.spellcheck.yaml
Fconstraints.txt
LICENSE
!mkdocs.yml
3.ATTEMPT-LLM triesto answer from the section text.
pyproject.toml
Returns (can_answer: bool, response: str}.
①README.md
OUTLINE
OCR text:
a
af
a ¥
a ’
ee ne ae ee ee ee ee ee ee rc a a ae ee ea
Pe a ee ee ee ee ee Re ee ee ee eee eee er
* .
ee Oe ee ee eee er me eae: Cera aa
pine ma eced cn eer POLL ea ay ia eae eae ree Beet ee a ee ee er a
a ae a ad RET SS TACAATUS ARS COLA To Ma ote re) os : i a De
ie Ee ge ee pec re ee ee eer ae ee
Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.