Ukuthi i-Anomalo ixazulula kanjani izindaba zekhwalithi yedatha engabonakali ukuletha izimpahla ezithembekile ze-AI nge-AWS

Lokhu okuthunyelwe kubhalwe ngokuhambisana noVicky Andonova noJonathan Karon abavela ku-AnoMalo.

I-Certative AI ivele ngokushesha yavela ngokushesha kusuka ku-Noventy kumshayeli onamandla wokuqamba. Ukusuka ekufingwiniwe kwemibhalo yezomthetho eyinkimbinkimbi yokufaka amandla abasizi abathuthukile abasenkingeni, amakhono e-AI ayanda ngesivinini esandayo. Ngenkathi amamodeli amakhulu wezilimi (i-LLMS) aqhubeka nokucindezela imingcele emisha, idatha yekhwalithi ihlala iyinto enqumayo ekufezeni umthelela wangempela womhlaba.

Ngonyaka owedlule, kwabonakala sengathi ukwahlukana okuyinhloko ezisekelweni ze-AI zokwakha ezikhiqizayo kungaba ngubani okwazi ukwakha noma ukusebenzisa imodeli enkulu. Kepha ngokuphumelela kwakamuva kwezindleko zokuqeqeshwa ngemodeli eyisisekelo (njenge-Deepseek-R1) kanye nokuthuthuka kwamanani okuqhubeka kwamanani, amamodeli anamandla aseba yimpahla. Impumelelo ku-generative ai izoba ngaphansi kokwakha imodeli efanelekile nokuthola okuthe xaxa ngokuthola icala lokusebenzisa elifanele. Ngenxa yalokhu, umngcele wokuncintisana uguqukela ekufinyeleleni kwedatha nekhwalithi yedatha.

Kule ndawo, amabhizinisi alungele ukugqama. Bane-Goldmine efihlekile yamashumi eminyaka yombhalo ongahleliwe – konke kusuka ku-Call Traterscript kanye nemibiko yokuskena ukusekela amathikithi nezingodo zemidiya yezenhlalo. Inselelo yindlela yokusebenzisa leyo datha. Ukuguqula amafayela angahleliwe, ukulondolozela ukuhambisana, kanye nokunciphisa izingqinamba zekhwalithi yedatha zonke ziba yizithiyo ezibucayi lapho inhlangano isuka kubashayeli bezindiza be-AI.

Kulokhu okuthunyelwe, sihlola ukuthi ungayisebenzisa kanjani i-AnoMalo nge-Amazon Web Services (AWS) AI kanye nokufunda ngomshini (ai / ml) kuya kuphrofayili, ukuqinisekisa, nokuhlanza amaqoqo wedatha angahleliwe ukuze uguqule umthombo othembekile wokukhiqiza ama-AI amahle we-AI, njengoba kukhonjisiwe kulokhu okulandelayo.

Inselelo: Ukuhlaziya amadokhumenti angamabhizinisi angahleliwe esikalini

Ngaphandle kokwamukelwa okusabalele kwe-AI, amaphrojekthi amaningi we-AI aphumelela ngenxa yekhwalithi yedatha engeyinhle nezilawuli ezinganele. UGartner ubikezela ukuthi ama-30% amaphrojekthi akhiqizwayo e-AI azoshiywa ngo-2025. Ngisho nezinhlangano eziqhutshwa kakhulu zedatha zigxile kakhulu ekusebenziseni idatha ehlelekile, zishiya amachibi angahleliwe noma amasistimu wefayela. Kodwa-ke, ngaphezu kwama-80% wedatha yamabhizinisi ayihlelwanga (ngokusho kwe-MIT Sloan School Research), ithatha konke kusuka kwizinkontileka zomthetho kanye nokuhlunga kwezezimali kuya kokuthunyelwe kwabezindaba.

Kwabaphathi bolwazi abakhulu (i-CIOS), izikhulu ezibucayi zobuchwepheshe (i-CTOS), kanye nezivumelwano zokuphepha kolwazi oluyinhloko (ama-cisos), idatha engahleliwe imelela ubungozi nethuba. Ngaphambi kokuthi usebenzise okuqukethwe okungahleliwe kuzicelo ze-AI zokwakha, kufanele ubhekane nalezi zingqinamba ezibalulekile ezibucayi:

• Isizinda – Ukuqashelwa komlingiswa we-Optical (OCR), i-parsing, kanye ne-metadata Generation kungathembeka uma kungenzeki futhi kuqinisekiswe. Ngaphezu kwalokho, uma ukukhishwa kungahambisani noma kungaphelele, kungaholela kwidatha engalungisiwe.
• Ukuhambisana nokuphepha – Ukusingatha imininingwane ekhonjwayo (PII) noma impahla yokuphathelene nengqondo (i-IP) ifuna ukuphatha kanzima, ikakhulukazi ngomthetho we-EU AI, iColorado AI umthetho, umthetho wobumfihlo wokuvikelwa kwedatha (i-CCPRISIA), kanye nemithethonqubo efanayo. Imininingwane ebucayi kungaba nzima ukukhomba embhalweni ongahleliwe, okuholela ekunciphiseni okungabonakali kwalolo lwazi.
• Ikhwalithi yedatha – Akuphelele, kwehlisiwe, okuphindwe kabili, okuphindwe kabili, noma idatha ebhalwe kahle kungangcolisa amamodeli wakho we-AI akhiqizayo kanye nokubuyisa umongo we-augment augmented (rag), ukuhoxa okuphumayo, okuphumayo. Ukuqinisekisa ukuthi idatha yakho ikhwalithi ephezulu isiza ukunciphisa lezi zingozi.
• Ukukala nezindleko – Ukuqeqeshwa noma amamodeli wokuhlelela okuhle kwimininingwane enomsindo kwandisa izindleko ezikhokhwayo ngokukhulisa ngokungadingekile idatha yokuqeqeshwa (ukucubungula kanye nokugcina idatha esezingeni eliphansi kwi-Votabase ye-RAG yokucubungula kanye nomthamo wokugcina.

Ngamafuphi, ama-adrantionitative AI amahle adonsela phansi ngoba imodeli engaphansi ayanele, kepha ngoba ipayipi elikhona ledatha awenzelwe ukucubungula idatha engahleliwe futhi isahlangabezana nevolumu ephezulu, izinga eliphakeme lokungenisa kanye nezidingo zokutholwa. Izinkampani eziningi zisezigabeni zokuqala zokubhekana nalezi zingqinamba futhi zibhekene nalezi zinkinga ezinqubweni zazo ezikhona:

• Ngesandla nesikhathi esidla isikhathi – Ukuhlaziywa kwamaqoqo amakhulu wemibhalo engahleliwe kuncike ekubuyekezweni kwesandla ngabasebenzi, ukudala izinqubo ezisebenzisa isikhathi ezibambezele amaphrojekthi.
• Iphutha – Ukubuyekezwa komuntu kutholakala kumaphutha nokungahambisani, okuholela ekukhishweni okungathandeki kwedatha ebucayi nokufakwa kwemininingwane engalungile.
• Izinsiza ezinamandla – Inqubo yokubuyekezwa kwencwadi yencwadi idinga isikhathi esibalulekile sokuthi ingachithwa kangcono emisebenzini yebhizinisi eliphakeme. Izabelomali azikwazi ukusekela izinga lokuhweba okudingekayo kumaqoqo we-Vet Enterprise.

Yize izinqubo zokuhlaziya zedokhumenti ezikhona zihlinzeka ngokuqonda okubalulekile, azisebenzi noma zinembile ngokwanele ukufeza izidingo zebhizinisi zanamuhla zokwenza izinqumo ngesikhathi. Izinhlangano zidinga isixazululo esingacubungula amavolumu amakhulu wedatha engahleliwe futhi asize ukugcina ukulandela imithethonqubo ngenkathi kuvikela imininingwane ebucayi.

Isixazululo: Indlela ye-Enterprise-grade yekhwalithi yedatha engahleliwe

I-ANOMALO isebenzisa isitaki esiphephe kakhulu, esilinganiselwe esinikezwe yi-AWS ongayisebenzisa ukuthola, ukuhlukanisa, nokubhekana nezinkinga zekhwalithi yedatha yedatha ngemininingwane engahleliwe. Lokhu kusiza amaqembu akho wedatha alethe izinhlelo zokusebenza ze-AI eziphakeme ngokushesha nangobungozi obuncane. Ukwakhiwa kwesixazululo sika-Anomalo kuboniswa kulokhu okulandelayo.

1. Ukufakwa okuzenzakalelayo kanye nokukhishwa kweMetadata – I-Anomalo automates OCR kanye nombhalo parsing amafayela we-PDF, izethulo ze-PowerPoint, kanye nemibhalo yamagama egcinwe insizakalo yesitoreji esilula (i-Amazon EC2) Sebenzisa insizakalo ye-Amazon Elastic Cloud (Amazon EC2), kanye ne-Amazon Eckastic), kanye ne-Amazon Ecr).
2. Ukuqashelwa kwedatha okuqhubekayo – U-Anomalo uhlola i-batch ngayinye yedatha ekhishwe, ukuthola ama-anomalies afana nombhalo oncishisiwe, amasimu angenalutho, nezimpinda ngaphambi kokuba idatha ifinyelele amamodeli akho. Kule nqubo, iqapha impilo yepayipi lakho elingahleliwe, i-passging egculisana ngamadokhumenti amaphutha noma i-Drift yedatha engajwayelekile (ngokwesibonelo, amafomethi amasha wefayela, inombolo yefayela elingalindelekile, noma izinguquko ngosayizi wedokhumenti). Ngalolu lwazi lubukeziwe futhi olubikwe ngu-AnoMalo, onjiniyela bakho bangachitha isikhathi esincane behlanganisa izingodo kanye nesikhathi esithe xaxa ukwenza kahle izici ze-AI, ngenkathi ama-cisos ethola ukubonakala ezingozini ezihlobene nedatha.
3. Ukuphatha nokuhambisana – Ukutholwa okwakhelwe ngaphakathi kanye nokuphoqelelwa kwenqubomgomo kusiza imaski noma ususe i-PII kanye nolimi oluhlukumezayo. Uma i-batch yemibhalo ehloliwe ifaka amakheli akho noma amadizayini okuphathelene, ingahlaba umkhosi wokubuyekezwa kwezomthetho noma kwezokuphepha – ukunciphisa ubungozi obulawulayo kanye nobuntu. Ungasebenzisa i-Anomalo ukuchaza izingqinamba zangokwezifiso kanye ne-metadata ukuthi ikhishwe kumadokhumenti ukuxazulula uhla olubanzi lokubusa kanye nezidingo zebhizinisi.
4. Skalible ai kuma-aw aw – I-Anomalo isebenzisa i-Amazon Bedrock ukunikeza amabhizinisi okukhethwa kukho okuguquguqukayo, ama-llms ahlelekile wokuhlaziya ikhwalithi yedokhumenti. Ukwakhiwa kwanamuhla kwe-Anomalo kungathunyelwa njengesoftware njengensizakalo (SAAS) noma nge-Amazon Virtual Cloud Cloud Cloud (i-Amazon VPC) ukuhlangabezana nezidingo zakho zokuphepha nezidingo zokusebenza.
5. Idatha ethembekile yezicelo zebhizinisi le-AI – Isendlalelo sedatha esiqinisekisiwe esinikezwe yi-ANOMALO ne-AWS glue isiza ukuqiniseka ukuthi okuhlanzekile kuphela, okuvunyiwe kugeleza kuhlelo lwakho lokusebenza.
6. Isekela ukwakhiwa kwakho kwe-AI – Noma ngabe usebenzisa ukufunwa okuhle noma ukuqhubeka kokuqeqeshwa kwangaphambili ku-LLM ukudala uchwepheshe wendaba, ngokuqiniseka kwezakhiwo ze-vector i-rag, noma uqinisekise ukuthi idatha yakho ihlanzekile futhi iqinisekisa ukukhishwa kohlelo lokusebenza, kanye nokunciphisa ibhizinisi.

Umphumela

Usebenzisa izinsizakalo ze-ANOMALO ne-AWS AI / ML zemininingwane engahleliwe ihlinzeka ngalezi zinzuzo:

• Kunciphise umthwalo wokusebenza – Imithetho ka-AnoMalo's off-the-Shelf Injini yokuzihlola Gcina izinyanga zesikhathi sokuthuthukisa kanye nokulondolozwa okuqhubekayo, isikhathi sokukhulula, isikhathi sokukhakhulula sokwakha izici ezintsha esikhundleni sokuthuthukisa imithetho yekhwalithi yedatha.
• Izindleko ezilungiselelwe – Ukuqeqesha i-LLMS namamodeli we-ML kwimininingwane esezingeni eliphansi – ngenkathi ukhetha i-veterting futhi ukugcina leyo datha ye-rag inyuka izindleko ezisebenzayo, kanye nobabili ukusebenza kwesicelo. Ukuhlunga kwedatha kusenesikhathi ukusika lezi zindleko ezifihliwe.
• Isikhathi esisheshayo sokuqonda – I-Anomalo ihlela ngokuzenzakalelayo kanye namalebula wombhalo ongahleliwe, enikeza ososayensi abacebile idatha yokuphoqa ama-prototypes amasha noma amadeshibhodi ngaphandle kokulebula okuthola ilebula.
• Ukuqinisa ukuhambisana nokuphepha – Ukuhlonza i-PII kanye nokulandela imithetho yokugcinwa kwedatha kwakhiwa epayipini, ukusekela izinqubomgomo zokuphepha kanye nokunciphisa ukulungiswa okudingekayo ekucwaningweni kwangaphandle.
• Dala inani eliqinile – Isimo se-AI esenziwe ngaso leso sikhathi siyaqhubeka ukuvela ngokushesha. Yize utshalomali lwe-LLM kanye nolwazi lwezakhiwo zohlelo lokusebenza zingahle zinciphise ngokushesha, idatha ethembekile futhi ekhethiwe iyinhloso eqinisekile engekeki.

Ukugcina

I-Certative ai inamandla okuletha inani elikhulu le-gartner kulinganisela ukukhuphuka kwemali engenayo engu-15- 20%, i-15% izindleko zokonga, kanye nokwenza ngcono okungu-22%. Ukufeza le miphumela, izinhlelo zakho zokusebenza kumele zakhiwe ngesisekelo se-ATTRETED, Qedela, ne-Timetely Data. Ngokuletha ikhambi elisebenziseka kalula, elinamabhizinisi okuqapha ikhwalithi ehlelekile futhi elingahleliwe, i-Anomalo ikusiza ukuletha amaphrojekthi amaningi we-AI ukukhiqiza ngokushesha ngenkathi uhlangabezana nazo zombili izidingo zakho zokuphatha kanye nokuphatha.

Unentshisekelo yokufunda kabanzi? Bheka isixazululo sekhwalithi yedatha ye-AnoMalo futhi ucele i-demo noma uxhumane nathi ukuthola ingxoxo ejulile yokuthi ungaqala kanjani noma ulinganise uhambo lwakho lwe-AI afanele.

Mayelana nababhali

Vicky Andonova Ingabe i-GM ye-ai ekhiqizwayo e-Anomalo, inkampani evuselelwa ngekhwalithi yedatha yebhizinisi. Njengelungu leqembu elisunguliwe, uVicky uchithe iminyaka eyisithupha edlule ngomshini wokufunda komshini we-Anomalo, uguqule amamodeli athuthukile e-AI ukuba aqonde kahle ama-Enterprises athembele idatha yawo. Njengamanje, uhola iqembu elingagcini ngokuletha imikhiqizo emisha ekhiqizayo ye-AI emakethe kodwa futhi yakha isixazululo sokuqapha sekhwalithi yokuqala esenzelwe imininingwane eklanyelwe imininingwane engahleliwe. Phambilini, e-InstaCart, uVicky wakha ipulatifomu yokuhlola yenkampani nezinhlelo zenkampani eholwa yinkampani yokulethwa kwezitolo. Ubambe ukuba aqhamuka e-Columbia University.

UJonathan Karon Ihola umucu omusha e-Anomalo. Usebenza kakhulu nezinkampani emgqonyeni wezembatho wokuhlanganisa ukuqapha kwekhwalithi yedatha kumathuluzi asemqoka kanye nokuhamba komsebenzi, ukusiza amabhizinisi kufinyelele imikhuba yedatha ephezulu kanye nokwakha ubuchwepheshe benoveli ngokushesha. Ngaphambi kwe-ANOMALO, uJonathan wadala ukubonwa kohlelo lokusebenza lweselula, ubuhlakani bedatha, kanye nemikhiqizo ye-devsecops e-New Relic, futhi wayeyinhloko yomkhiqizo ekuthengisweni kwe-AI ekhiqizayo kanye nokuqala kwamakhasimende. Ubambe i-BA eyisayensi yengqondo evela eHampshire College futhi isebenze nobuchwepheshe bokuhlola i-AI ne-DATA yonke yonke imisebenzi yakhe.

UMahesh Bhiladar Ingabe ukwakhiwa kwezixazululo eziphezulu ku-AWS ngomlando embonini ye-IT kanye nezinsizakalo. Usiza ama-SBB e-US ahlangabezana nezinhloso zawo zebhizinisi ngobuchwepheshe befu. Ubambe i-bachelor of the Engineering kusuka ku-VJTI futhi isuselwe eNew York City (US)

I-Emad Tawfik Ingabe ukwakhiwa kwezixazululo eziphezulu ezinamava ezinsizakalweni zeWebhu zase-Amazon, ukuziqhayisa ngaphezu kweshumi leminyaka. Ubuchwepheshe bakhe busendaweni yokugcina izixazululo zokugcina kanye nefu, lapho edlula khona ekwakhiweni kwezakhiwo ezingenazindleko ezingenazimali nezingahleleki kumakhasimende.