Dense Slides: Ship Real Agents: Hands-On Evals for Agentic Applications — Laurie Voss, Arize

Source Video

Ship Real Agents: Hands-On Evals for Agentic Applications — Laurie Voss, Arize

Method

This deck is slide-only. The existing captured video frame set supplies candidate frames, then local OpenCV rejects sponsor/title/speaker-only frames, crops visible slide surfaces, deduplicates, and saves the cropped slide images.

Cropped Visible Slides

Recreated text/layout view: open HTML recreation
AI slide classifier: title_card confidence 0.98
Text source: agent_vision.

Slide text:

Get set up

1. Learner notebook: bit.ly/build-real-agents

2. Claude API key: platform.claude.com/settings/keys

3. Phoenix Cloud account: app.phoenix.arize.com/login/sign-up

• Phoenix hostname: app.phoenix.arize.com/s/yourusername

• Phoenix API key: app.phoenix.arize.com/s/yourusername/settings/general

Recreated text/layout view: open HTML recreation
AI slide classifier: title_card confidence 0.99
Text source: agent_vision.

Slide text:

What you can't do without evals

• Can't detect regressions when you change a prompt

• Can't compare prompt versions objectively

• Can't know if a new model is actually better

• Can't run quality gates in CI

Recreated text/layout view: open HTML recreation
AI slide classifier: other confidence 0.98
Text source: advanced OCR rapidocr-live/bright-screen/contrast.
OCR decision: ready — Dense code-style explanation text under a large title.

Slide text:

What a real explanation

looks like

label: "incorrect"

breakdown, which is a core requirement. The agent explanation: "rhe response fails to include a budget

but omits all cost estimates, making the plan incomplete for a user who asked specifically about budget travel to Tokyo." provides destination info and local recommendations

Recreated text/layout view: open HTML recreation
AI slide classifier: other confidence 0.97
Text source: advanced OCR rapidocr-live/border-trim/opencv-adaptive.
OCR decision: ready — Dense product UI screenshot with many small labels and values.

Slide text:

PHOENI Trsce 1 10 e5c4131942643c221s01360s654a5: G

...... trant tot!. O UHSEY Tott Cot. Uorey 01 161 Oé. cortettt O:.8 OF1.68 ○.8 T'iet Yehhy.

. <1. NxT AH ① onoial: Purtund Powga':. LL 1 s weesrch + A wotseece a wotoca Lpart: + 1D Ja221b341r87c7: G 1a 164 m 0/22/2024 09:08.42 P4:os. fnnctLreport tite A++rch: TsLA Focu+! rthtele dellr!it ad rph √inps Arnotdons:4),. Evend t. +'i Luet = Lekt Aanoteor I mtoe,tear me. h

9 +osrdh output +Aro.xn

[ttalrch fledirpa lato a ptofestaal dotmet.,. The t+port. Includts! ① ""firanoinl reoort crated succecsfuily): 1++ aritt+a: concit+ rirncilt' ieport +a I+ela_that tmth+slr++ tha '+tta + 1m 't1 ts 1u e 1rdormte tr 1h. Seact M tene

Hro vo Ag U tkey Finrcial Mttlce Table. - To-ovei-ynt <o.prltors sharing $.ot:-..+"bouiros Porforsice Aealraie+'. - Cetalled bioshson ot: - t+gtrateole Iaitistive+' - pecord Captx ellocatieo, Tesla Stal piotuetion; o: td oirot to tuteroeout dtlrlr/robotic4..::. C+llweries, mipin co=proioa) sd rtategio pltot:..':::. coetatlol res)...rr tocttema".. dol1wery d+clir+: 1a.t34 tto+4 rg1a. td 1ecord 12+e+ CacEt pla "Exrcutive s Yhicle doltvety detertoratlon (1.em units, tecons etrsight drclln) "Ank Faetore+ - Ieclusirg sriet wtulity. coc+tltton, s4 otorttoa'. - Crotu Mtoi+ coilroae to 14.e1s frcn 2s1+ p+aka:.. = (ell deelire of J?.6l dtite ret 1acone drorlh (hlghllgntiro - C+rrle tt Tela'+ ohllta (t+olinirg.* ：...

tote: 安 tupeort eietntation.: Letitutional-grsde arelysie witeble fei 1reetmont comalttees ol board +2e2s outloe** - Ouidhnc+ sh+ed er Aortl 21 C1 4arningu cal) Tm I+oort 1g tve oe +I+sli.firac1s1 h+cort.224.ag'- sed pror1t+t *Concluaioo'+ - Ascetaent of Tealn'e ballrett medel traasitloo. ·..

1+ Lsg Gut Add $ Ao.

Recreated text/layout view: open HTML recreation
AI slide classifier: other confidence 0.96
Text source: advanced OCR rapidocr-live/border-trim/opencv-adaptive.
OCR decision: ready — Diagram with small labels and annotation text.

Slide text:

The Swiss Cheese Model

Zooming Out: The Swiss Cheese of Quality.

Auromaledevaier

P:ovidn conehlent mamtremant.

keulodep uo,q tuosso dhy benchearki, sd cetchet:

Mencn smoript rerew,

Unexpected usar or scerkt beharvior Cetchtt rusnctd fahuret thd

Preducten menltorng. surt sees rere edge cesee esd rth utare pattarr+ t tetle A's t+eting utr t++otack!

Recreated text/layout view: open HTML recreation
AI slide classifier: title_card confidence 0.99
Text source: agent_vision.

Slide text:

Part 2: Explicit criteria

• ACTIONABLE — The report:

▪ Contains specific recommendations (buy/sell/hold or equivalent)

▪ Identifies concrete risks with supporting data

▪ Includes forward-looking analysis, not just historical data

▪ Provides context for why recommendations are made

• NOT ACTIONABLE — The report:

▪ Only summarizes publicly available data without interpretation

▪ Lacks specific recommendations or next steps

▪ Presents risks without supporting evidence

▪ Contains only backward-looking analysis

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.99
Text source: agent_vision.

Slide text:

What makes criteria specific

ACTIONABLE — The report:

Contains specific recommendations (buy/sell/hold or equivalent)

Identifies concrete risks with supporting data

Includes forward-looking analysis, not just historical data

Provides context for why recommendations are made

Classification audit: raw/sources/slide-ai-classification/dense/Xfl50508LZM/audit.json