--- title: "Dense Slides: Building Intelligent Research Agents with Manus - Ivan Leo, Manus AI (now Meta Superintelligence)" category: "slides" video_id: "xz0-brt56L8" sourceLabels: ["Captured video frames", "Local OpenCV slide-region detection"] --- # Dense Slides: Building Intelligent Research Agents with Manus - Ivan Leo, Manus AI (now Meta Superintelligence) ## Source Video [Building Intelligent Research Agents with Manus - Ivan Leo, Manus AI (now Meta Superintelligence)](https://www.youtube.com/watch?v=xz0-brt56L8) ## 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 ![[assets/dense-slides/xz0-brt56L8/slide-001.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-001.html) - AI slide classifier: `content_slide` confidence `0.99` - Text source: agent_vision. Slide text: > 4x faster. 15% better. Production-ready. > 4x FASTER > 15% QUALITY GAIN > 6% USER SATISFACTION ![[assets/dense-slides/xz0-brt56L8/slide-002.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-002.html) - AI slide classifier: `content_slide` confidence `0.88` - Text source: advanced OCR `rapidocr-live/bright-screen/opencv-adaptive`. - OCR decision: ready — Dense UI screenshot with small body text and multiple cards; OCR will likely read this more reliably than manual transcription. Slide text: > ? Cort > Wodnesdsy, Novenbor 19, 2025 > 1.06 Pu > Li dta rt que Lrt.el yi elina toaed mo ga dett cosnernertrart T Hehsha Tu e acrs (nouud ge anai Lao Lraa pue itatg pln hcly duu satey dupar Lo mnuoy > d Shou Corricted Hertitn > Y't'tor tooiltence. Focust cn verb conlugatian prlintht srher modtr verht, corier hmse, ane iproring common eipror: Pae Y > Aioalu 't dahla mon biml C metaryt parce ou'Lraas nonbatuueukdhy I srcongr. Sepon sccompkr toutes les tlchee srent roon dtnet ce coir -- > oYhou Corltitd Hrtor > Imd chooing mone nin, ihomrat phnryng. The srun srhs ior hmproyerert are prhpoulian ussge (h vt caal. corrtet vtrb conuguior (otr+ corpou!. > Aujeurhu a teal trop frold gses Nre Yet. Je wit Hu bureu lgot de Fuet ron pour tire mon tarel Je oerche 99260069560 C.tl- *2025-11-22*09:37:50* ![[assets/dense-slides/xz0-brt56L8/slide-003.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-003.html) - AI slide classifier: `content_slide` confidence `0.93` - Text source: agent_vision. Slide text: > Long URL > TinyURL Link ![[assets/dense-slides/xz0-brt56L8/slide-004.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-004.html) - AI slide classifier: `content_slide` confidence `0.98` - Text source: advanced OCR `rapidocr-live/bright-screen/opencv-adaptive`. - OCR decision: ready — Multi-column editor/notebook screenshot with dense small text and code-like content. Slide text: > 。1 +t i ttt 114 P地 + t+tt. 1 *Lo* + Cth + Lia - O LO lt +1. C te LL + T Ae.:? Creating your first Manus Task Wetcorme io tre Drst ste in your Mne joneyi re netecoct is st sbeul peting rou handt-on wal iye Ail tm guidlly ae Dosltia, hed the the lorg iecitree erd yirg igM Tts tiu col miriur'ice c cresirg and rlruyig LaLn trog ttshr eby IPeaLnar Sign up for ar sccount mt Tnge,rLnL nd cla yur Tru At tet woulg tt' g 1 Manus APl Workshop Hlcora to te thrue APt wcna.ip. That tapotiaory brcd + touetre, h+t 1. Lh fcdalenthr Th, hetebooh colnt it cole rhthhiet et the Larus APL Tl im tm hotr to sartncCia, Crtt Letat, po lor tht mHLt, snd > 目1 W*2*** *4b sulonhhot. Tere's e ouict took et ithe: t'lt coitr. Gy te thd at aha mheor, youl hre + hoha taurclPon iar buittlg shae toced C ieh,suru, + walj mrs e iaa h aou * ram alaen ytei op atri t 4. Leverige Ceamrelny we' rnmh eh + aukh look a hew you Can ga Llan 2. Crele Teur Firt Thl: fou't ghe Larlst + wrele reructon ard lerh hor t A ar ia, + Dunna pu+ Lir rol Dupto Aa Let L+t:eunt y+hoey ' Tttrpil tle h+i ihoorh. 4oCth lo yot eit tohy, ile Hoear, te ptriarln mere corlcleu iore folt. '+yrpor so iay +a ctod Aa hohuto, sa,Irpu+d. 2. Fha Harelrg Heet yot'a trt hou to hart te hhet ih Ia Lur.s Al to ithit 1. Hebr:olt: Pohrg do+a't hent s tktle L+ra try ohcol sre + utsill Yaur sott hae ir+ da I rteoe to fhnon electyef. 11. > Lr's oat +hted. Settinn IIn Yaur Fnvirnnment L't 'Puprcot aohovoo +a se ho 1+os rus ell aod Loddg t tuaper $ Cld e deglojiote tlpport sotnt tht ctn practe Da yetotdt trem stct tnd ToBEL > # +- n - A > Tttett, 'ttur Cl?t 'o tt Lh LF! > t+++ +u *+ Pa *.++'! ++ ++ Or!' tta. ++ + ovt+ L+t 4 +tt +ttta > t Cenciomen ++ Cartlla + tt 11-401m > 2..1 2025-11-22:09:49:28 ![[assets/dense-slides/xz0-brt56L8/slide-005.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-005.html) - AI slide classifier: `content_slide` confidence `0.97` - Text source: advanced OCR `rapidocr-live/full`. - OCR decision: ready — Code editor screenshot with small source text and file tree; OCR is the right capture method. Slide text: > dgg tk.y sra 19 29 日日 2t 14 det Loae Artey frasenefile sat_sotetvl) L441 *4164x firaL1d > md 23 ty ai_kty-es.gten > -oUTLIN 2.Finiy qT READMEmd y op Botmd gtge WLne 13 34 35 31 2 60 《1 K2. <3 5 ) $9 54 Tet te R y w the files odpeint Valiate e ey i loeo ftaL_ky1 tryi eetepet.ecgt et tnae respoaeregest-ttfrt,Maere etr eed rle eltr"4 iFe.resose.stat_codeel else! ralse xalotrrerf ra YsluttrrrA ratsetrorr RsTTrrr aset PLeet Taolef te valssete Adt te Chel mieuy Dxtla(+(3tr[e)2*) (strlel)) > s1 51 def pet_hae_rli)! > 3I71 > 19 returs > 2025-11-2209:57:30 ![[assets/dense-slides/xz0-brt56L8/slide-006.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-006.html) - AI slide classifier: `content_slide` confidence `0.97` - Text source: advanced OCR `rapidocr-live/bright-screen/opencv-adaptive`. - OCR decision: ready — Dense notebook/code screenshot with multi-pane text and JSON content; OCR is appropriate. Slide text: > 0 1. Fht + 1 F+ l m a i + h t P. Ah + + 1t e + + + 9t? *+ G: D + f tom''pu + 1? > oraL Lan, + 鲁 4 tad * +S +LD. @ % Y ++ fe Heheg (t: +_M + tkr + ( ro m metharwry) Tal TnL ie mal to Lplotd a tie te gt a tecLre, olLit Lcioal Lir. (toord, ite thd te san tle Tilh cur leet Ple rehuy, ir't tire lo upiotd x to Hanuel. Th+ procts irmche hao tirrpte tegts. Prct, mt Lth Ta tt LIr ++i + t..rini+ +ile reh+ ho+r!..att=l · fil_kiI'olem. rtste_Pila_renommoe eisr_rer_rtatat!. 1*rn+hta. Crete_fite_rewert 4'rsorell.pit] at+ -t?ilet,.+rI t! + h+ lr lees Ie flle ctet, ro, ra m: sta i Crote 's tie rtcove is tt i paiges uiom se. I.e.ieouuti + pr nt, tIle_rote = (rrte flla_re4./--)': Pmr',: *p+++ rt.e + r+e,t+.ptt' -vt_ari ≥iar, l++ re+eae relt_ter,statust] mleLrl.. Huts: b. B = n 7 + u 40 77 n 11 14 11 L 11 17 11 n 74 # L 11 11 1 1 ， 1 "y', ". "h+hr'+ *++?-1-+?u++ +.?. *1 i.M- '.-. I.+t. "mtr! 'eir. "u'!?. "a+'! 'Tury e.ih'. *.+. I.u+!. "' "+uu"! 1 "ttrtu'+ "llt. "++l+'1 1 1 t,+++ ++. -1 1,*1a, 1 1tM** 1. ++++'! "(+++ol ++ +++.+. "++'+ -+.rh (c-im=. --I.14, +.++++ a +++++t. t.-+. > [4]. 2a +Fiy plen sxorellipi + Tite rerere cerale4 P+h Dt fil-xerrphpru (.++taytu+ h+iy+ rt) /.i hot..r vtir?le t tr.1eh nttm!+'(hl_ho): 1 1 + t u +1 11 41 ""t! 1, > 9BA801069 22 1* + 1 2025-11:22 09:58:45 ![[assets/dense-slides/xz0-brt56L8/slide-007.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-007.html) - AI slide classifier: `content_slide` confidence `0.83` - Text source: agent_vision. - OCR decision: ready — Product/UI screenshot with dense small labels, charts, and side navigation; OCR will likely capture more accurately than manual transcription. Slide text: > Rick and Morty Character Dashboard ![[assets/dense-slides/xz0-brt56L8/slide-008.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-008.html) - AI slide classifier: `content_slide` confidence `0.95` - Text source: agent_vision. - OCR decision: ready — Dense code/notebook/editor slide with small text and multiple panes; OCR is more appropriate than direct transcription. Slide text: > Conclusion ![[assets/dense-slides/xz0-brt56L8/slide-009.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-009.html) - AI slide classifier: `content_slide` confidence `0.94` - Text source: advanced OCR `rapidocr-live/full`. - OCR decision: ready — Dense code slide with terminal and editor panes; OCR will recover small code and log text better than manual transcription. Slide text: > VOUTLME gtigrors 1Whos.pyb quanprs18 5pBemd 2.F MeWb 0 dbgong mPy nk.Ey )md 1L deta 中 nicinmotyc manscrietsoe vsrtnet ocuents/ceding/mn-api-ethoo-rkshoivans（a]sorce,vem/bn/actiate +Code+Makdown [A3 (ad-api-ebhso-erkshop）/Documents/coding/anus-api-ehok-orkshop Lvans（nain}odal serve server-py /uiers/lanlaoor/bocents/coding/ni-agi-ehoo-ehog/ven/ia/pytho.9/site-gackage/arlLib3/_init Protor varings.ware senver endpoint wlthout any issues. 10 saperr regests Send tet rest Cratest aylad tet resonsereeest.ast（ Lapert test_paytsat HCDAL_O_L, walerpa( Jsnitest_Daytast RAC Perts 日 venv (Python 3.9.6) we can hit our npor1rt sny 23 23 21 27 29 >E X3 H 35 34 X7 31 4F 43 43 41 45 Py135:Not0g Check if the tskh fishee fent_tyet_stp hk(payt： it): irtask_isi task_detail-payled.getitec_rai1²,（1) tatk_1d-task_oetall.gett²（ao_i) pitira feth the foll tatk ject fre te Ms AP irrsponse.status_code =2ae: etse: py-i.eir hsurl- respos+repetts.gt( stifutLtaortails) fell_tas_detaile·response.jh(1 pitllaal] sutirx TXZIEY SSL1.1.1,cufretly the'ss1'nodue Syh > Createdobjects. 051/（rt346.4s,t223.31） NeceLadeveot： mege_moed PO5T/（4,5.:219.） Receivedrventimessage.eided -// Createdsount/ers/ianleoeck/Doets/codig/nnt-apj-bhook-orsho/serer-py g/1c...h Serviag...hitCtrl-Cts > 2025-11-2210:07:30 ![[assets/dense-slides/xz0-brt56L8/slide-010.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-010.html) - AI slide classifier: `content_slide` confidence `0.97` - Text source: advanced OCR `rapidocr-live/bright-screen/opencv-adaptive`. - OCR decision: ready — Text-heavy code and log slide with small paragraphs and console output; OCR is the right extraction method. Slide text: > + ttrt: 糖 1 ahton +++ + T th+ar + PI$+g Yu 'tt+o + + + C + e L t c(+ a b + (I tH n?! Tls' oper stbarn, trom L:Lk crmton to sytrt ronfcton, i wirtut reyrg on polrg. r+ $yt 1.1.4 > L! ta+ert+ Processing Events > 1 +y. + t.+ 中 Te cet the Lurve Uh iromm tur Uoch itrctier, e hed our Lt ksy, Hs thocd hrle herdeode tartt. phtyt seutb ybe ta te Yuet pue tce+l! T ou Codt. Todt protds + ftert tey t Prala tiie ihg od Stael P+crirg + ebhool, me+retlen ie +e ft yt+e. The tee po+ comrt thont shst you do widh l. T? Pun tt iotoho corhila 'n rour Wrn'el, rt crthtt a tert Atd rtrct -tptty tt Gur torig Gr hh' Iath. > 静 Fls 4y-b tol+! Htrat 口 Tht gre sthe lolotvg cutoct > 量 4. pactLtry动 +tCiatk g? I & Apn toa (tatt + rr Hecret 'mres-pl-ley' +it tte Ley 'taus_Al_!y' (+u+-+pi -+rtho+h -+a++ Lhe+) +/b+o-.t U(ed Lr/taAu+-+e!-++taak -++-ithop Ireh > U++ tl in your todal #p+! > (+ > ￥... > (+* +*+r: h A*a *a t+h c+a or + / 1o4 +l+t emt! t,tahth! > U t torewatraaoult toe. Arrtp tall ttnlo... X rt trtty lhl tiu+ h: + S+all Uih rtulp! Pit /'+ 2a oa. (eatrr 4).+ eo. p*co 4 Ht.] ai trctl4 tatt 1a,+t4 M411 I'sw't.'wt+ 1*.] 1.=o). -1.-N tert* Ttt m clentte tn or, veu,. t'u'+hayrird! *'0 art t? + 7? Ie hr + +it. > 9四@0106@ Crh i I=+ C+ y tn +t 2025-11-22*10:10:25 ![[assets/dense-slides/xz0-brt56L8/slide-011.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-011.html) - AI slide classifier: `content_slide` confidence `0.96` - Text source: advanced OCR `rapidocr-live/full`. - OCR decision: ready — Dense instructional slide with code and logs; OCR is more appropriate than direct transcription. Slide text: > k.ry +empy data Frr iindasreet,yoave yr taseRL.Th pls the eneint path that we sefine erlier. Ly care,y fts 十Markdown 0let-s-5tk-sms-briege-fste]-0g-eee.msst.roe/weheks/stac nAer CAOs I96E 90 11 3 H 14 Sefim Ie-secret reetee intle arevjao ife Ypeats", > .Spt omd ch.pi RDADMEn 1AVixoetpin 9. 1md uwloo ect.soml o-time Rsretithtye eurverificatiotdcallene string. Sefore Slack send a reslevests,tseferesecrity hndske.It sends Your server ust resgne vith tat eact challenge string tereve it'sListeng. verityng tereeestsigntre teeere eeeg Jepertsedai ipertrequtsts apert json froe slackspk.sipatre iaprt Signatoreverfier Dmape-eodaL.inage.debi_sUnti.pia_sstai1 isert tastapi lAS repoest isgensirely fron 5lsck. 11 2# 21 22 2) 24 25 2$ 27 71 3# 31 12 33 14 sck_secretl.ecet.femfa caeg,tonetien( iaogeitage,secrets*[slac_secret1) cnodal.agt_get) _a-fst.FaP11) jap3ate retmvo_! retare (° eppt ILT t te Slack 1: Tastapi.Regatstl: > -OUTLINE g ac p 1 w Depc ME AGS cg aRegly igFaes ng our Senver dpg Slack.r Sack lhtegrati Steging ap-easare nterrpt receee.ueaodstru (n-jebhoo-erkshog）Docuaget/coding/-pi--orhogvn()darecatet.py Ar// /users/lvnte-won/Bocusnt/coding/aanu-ap2-ebhook-erkshog/.vene/La/pythan3.9/site-package/arllib3/_snit_Py:35:otpe cpiethUares2.L.]'.See:ts://git.c/artua/orLa/se/820 Created objects. amings.ern( oledeventua_tat P5/2009（38.m224） terelvedtveotr tiiatonped rroftchung Task etattsi 44 Crestedsout/ers/ivenlee-or/Docuents/cding/maus-api- 2aGzkaasis.etoi fs.. oupit Oebug Consolr drtahtegreent yoraterrptsfros iling as. > POSTMNeL/ack POST /ethocks/sSack watchsng/Dsers/jeanlee-werk/Docamests/ceding/sanus-api-thoek-workshep. / 0 200 (aurgUoni (doratisn) 5.21. 35.98. eecutiani 218.4 > 2025-11-2210:14:10 ![[assets/dense-slides/xz0-brt56L8/slide-012.jpg]] - Recreated text/layout view: [open HTML recreation](/assets/slide-recreations/dense/xz0-brt56L8/slide-012.html) - AI slide classifier: `content_slide` confidence `0.99` - Text source: agent_vision. Slide text: > Areas for further exploration > Advanced agent patterns: multi-step reasoning, tool composition, and error recovery > Production deployment: scaling, monitoring, logging, and cost optimization > Custom integrations: connecting to proprietary systems and internal APIs > Agent orchestration: coordinating multiple agents for complex workflows > Specialized domains: legal document processing, medical data analysis, financial modeling > Performance tuning: optimizing latency, throughput, and token usage Classification audit: `raw/sources/slide-ai-classification/dense/xz0-brt56L8/audit.json`