Chinyorwa-ku-SQL Chaifanirwa Kuve AI's Killer App. Hazvisizvo.

Mukupisa kwekutanga ChatGPT kupenga, ndakawana mameseji kubva kune aimbova wandaishanda naye. Aida kumhanya zano neni. Nguva yese munhu arikunakirwa nebrainstorming, we hopped on a call and he started off with “Remember how you used to always ask me to pull data for you? Ko dai wangozviita wega?" Uye ipapo anoenderera mberi nekundipa zano rekuti zviuru (makumi ezviuru?) Zvevamwe vanhu vaifunga panguva imwe chete: maLLM anogona kushandiswa kunyorwa-ku-SQL kubatsira vashoma tekinoroji vanhu kupindura mibvunzo yavo yedata.

Ndakabatikana nepfungwa iyi, asi ndisati ndanyura mumusoro kutanga, ndakaudza Lei (zvino CTO yangu) kuti taifanira kuita kumwe kusimbiswa. Takataura neshamwari uye vataimboshanda navo kubva mumaindasitiri akasiyana-siyana. Paiva nechido chakasimba mu "self-service analytics" chaiyo. Taiziva kuti zvaizonyanya kuomarara kupfuura zvazvairatidzika, asi mukana wacho wakanzwa wakanyanya kunaka kuti urege. Saka ini naLei takabva paShire tikatanga rwendo rwedu rwekugadzira chiono chedu: Fabi.ai .

Iyi positi haisi yechigadzirwa chedu pachayo (zvisinei, kana iwe uchida kuziva, unogona kuverenga zvakawanda nezve mamwe mazano ari pazasi akazivisa basa redu rekupedzisira chigadzirwa. here ) Pane kudaro, ndaida kugovera zvidzidzo zvekutanga zvatakaunganidza kubva mukushanda neLLMs yekuongorora data parwendo rwedu.

Ongorora: Rwendo urwu rwuri kushomeka zvinosiririsa muvaroyi uye epic Middle-earth hondo. 🧙

Sei uchifanira kushandisa AI yekuzvipa-sevhisi analytics?

Hatizorambi tichiti “nei” nguva refu. Kana uri kuverenga izvi, ungangowira mune rimwe remapoka maviri:

1. Iwe uri mumwe munhu anoshuva kuti iwe uine wega-sevhisi analytics iripo uye haudi kugara wakamirira pane yako data timu.
2. Iwe uri pachikwata chedata uye wanga uchinzwa nezve mashandisiro AI ichakubatsira kugadzirisa dambudziko rako rekukumbira ad hoc.

Kusava nehanya nebasa revanoongorora data uye masayendisiti, pfungwa yeAI inoziva zvese inogona kupindura chero mibvunzo nezve data yesangano inonzwika zvakanaka. Kana zvirinani, zvinonzwika zvakanaka kusangano uye vatungamiriri varo vebhizinesi vane hunyanzvi hwekuita nzira nyowani dzekubvunza mibvunzo hapana chinosungirwa. Iyi AI inogona kunge iri mhinduro yekugadzira "data-inotyairwa" sangano apo mutungamiri wese anotsamira pane humbowo hwesimba kuita sarudzo dzavo. Uye zvese pachidimbu chemutengo wazvaizowanzotora. Pakupedzisira! Masangano anogona kuita mari pa "mafuta matsva" avanga vachinzwa nezvawo kubva 2010.

Asi kana iri riri dambudziko rakakosha kugadzirisa uye AI yave yakanaka kudaro, nei pasina chigadzirwa chakanyatso kuigadzirisa kusvika pari zvino?

Nei AI yekuzviitira-sevhisi analytics yakundikana kusvika parizvino

Ongororo dzichangoburwa dzeindasitiri dzinopenda mufananidzo wakaoma wekutorwa kweAI mubhizinesi. 61 muzana yemakambani vari kuedza maAI agents. Zvisinei, vakawanda vanonetseka nezvekuvimbika uye kuchengeteka. Muchokwadi, 21% yemasangano haaashandise zvachose. Kuzengurira uku kunonzwika zvakanyanya mukati mezvikwata zve data, uko kurongeka uye kuvimbika kuri mishoni yakakosha pakugona kwedu kuita basa.

MaAdopters eAI-kunyanya mubhizinesi-ane bhawa yakakwira kana zvasvika kune zvinotarisirwa tekinoroji. Muchirevo che data analytics uye yekuzvishandira kurota, isu tinotarisira yedu AI tooling ku:

1. Inopa manzwisisiro: Matafura nemachati akanaka, asi aya ndiwo madiki ezvinganzi "njere". Maonero anoti "Aha!" nguva dzinouya kubva pakuona zvinhu mudata rako izvo zvinomhanyisa kune yako intuition uye ingadai isina kufunga neimwe nzira. Dzimwe nguva mubvunzo weSQL kana pivot inogona kuvhenekera mwenje pane izvi nzwisiso, asi kazhinji inonzwa zvakanyanya sekuwana tsono muhata.
2. Shanda zvakavimbika ingangoita 100% yenguva: Chinhu chega chakaipisisa kupfuura kusave nedata rakashata data. Kana iyo AI isingavimbike kana kufungidzira mhinduro uye data, izvo zvinoperetera nhau dzakaipa kumunhu wese. Izvi zvinoreva kuti kana AI iine data, inofanirwa kuishandisa nemazvo. Asi kana ikashaya data, inofanirwa kudzivirira kupa mhinduro (chimwe chinhu maLLM chine mukurumbira wakashata).
3. Svika kune akasiyana siyana ehunyanzvi hunyanzvi seti: Runako rweLLMs nderekuti unogona kutaurirana navo sezvaungaita nemumwe waunoshanda naye pamusoro peSlack. Unogona kushandisa mutauro usina kujeka. Mumwe munhu kana chinhu chinogona kunzwisisa chikumbiro chako mumamiriro ebhizinesi. Sezvineiwo, iyo yakawanda sisitimu inoda kushandisa chaiwo mazwi mune chaiyo fomu, iyo isingasvikike. Rudzi urwu rwehurongwa hunoda kudzidziswa uye kusimbiswa, izvo zvatinoziva tese, zvinogona kuve zvakaoma.

Zvinosuruvarisa, mhinduro zhinji dzazvino dzinoshandisa chinyakare monolithic AI framework, iyo inowanzotadza kuzadzisa zvinotarisirwa. Mumakore mashoma apfuura, boka reFabi.ai neni takashanda nesimba panyaya iyi. Isu takavaka prototypes yebhizinesi uye takaongorora akawanda sarudzo. Mukupedzisira, takaona kuti kana Retrieval Augment Generation (RAG) kana kunyatso-tuning yaigona kugadzirisa dambudziko iri neino monolithic framework.

Patakaedza nzira iyi, zvinhu zvishoma zvakava pachena kwatiri:

• RAG haina kusimba. Yakanyanya kudiki mamiriro uye iyo AI haigone kupindura mubvunzo uye njodzi yekunyepedzera. Yakawandisa mamiriro uye iyo AI inovhiringika uye inorasikirwa nemazvo.
• Imwe-pfuti AI haina kwainosvika kwauri. Iyo yakanakisa AI munyika haizombo kwanisa kunyatso kudhonza uye kuongorora data mupfuti imwe chete. Pane angori akawandisa nuances mu data nemubvunzo. Ngatitorei muenzaniso wakapfava unobvira: Une “Account type” ndima ine 95% ine vanhu gumi vakasiyana. Kana iwe ukakumbira iyo AI kusefa pane seti yeakaundi mhando, inogona kutadza kuziva kuti pane isina chinhu kukosha, nokudaro ichigadzira iyo isiriyo SQL mubvunzo. "Chokwadi," ungati, "asi isu tinogona kungoverengera zviverengero zvendima yega yega uye sampuli tsika uye tochengeta izvo mune yedu mamiriro vector chitoro." Mhando dzenyaya dzinenge dzisingaperi uye dzese dzakasiyana nenzira yadzo.
• Data yebhizinesi yakashata. Izvi zvine chekuita nepfungwa mbiri dzekutanga, asi zvakakodzera kusimbiswa. Kunyangwe kana kwenguva pfupi, sangano rinogona kuve nematafura mashoma egoridhe ane yakanyatsotsanangurwa semantic layer, zvese zvinouya zvichipunzika kana mutungamiri weRevOps afunga kugadzirisa bhizinesi modhi. Ndinoda kudhirowa mufananidzo wemba: Unogona kugara uchichengeta imba yakatsvinda, asi panogara paine chimwe chinhu chinoda kucheneswa kana kugadziriswa.
• Mavara-ku-SQL ane miganhu zvakanyanya. Kune yakawanda data analytics mibvunzo, kunyora iyo SQL kudhonza iyo data ingori nhanho yekutanga. Iri ndiro danho raunofanirwa kutora usati watombotanga kubvunza mibvunzo inonakidza. SQL haingogone kubata ongororo yakaoma iyo vashandisi vebhizinesi vari kubvunza. LLMs nePython kune rumwe rutivi dzakanyatsokodzera basa racho. Zvishandiso izvi zvinogona kutora yako SQL kuburitsa uye kuwana iyo tsono muhata. Vanogona zvakare kumhanyisa regression ongororo kuti vafukure akakura maitiro.

Mushure mekutarisa nyaya idzi, takafunga nezve maitiro ekuita AI kugadzirisa zvirinani kumatambudziko. Ndipo pakapinda maAI vamiririri uye vakasimbisa iyi pfungwa kwatiri.

Ramangwana: Agent meshes

Iyo miniti yatakaisa maziso kune agentic frameworks, takaziva kuti yaizochinja mutambo. Takaerekana tanzwa kuti taigona kurega AI ichifunga mapinduriro emibvunzo. Inogona kushanda nematanho uye kugadzirisa yega. Kana iyo AI ikanyora SQL query inopotsa nhando mu "Account type" ndima, inogona kuomesa-mubvunzo, kuona kukanganisa, uye kugadzirisa iyo pachayo. Asi ko kana tikakwanisa kutora danho iri mberi torega iyo AI ichishanda zvakanyanya muPython uye kuwedzera maLLM? Zvino, iyo AI inoita zvinopfuura kudhonza data. Inogona kushandisa Python mapakeji kana maLLM kuti uwane kunze, mafambiro, kana yakasarudzika njere, iyo yaunowanzo tsvaga pamaoko.

Asi isu takanga tichine dambudziko rimwe chete: iyo yakashata bhizinesi data. Isu taitenda kuti masangano anogona kugadzirisa izvi nekushandisa yakasimba data engineering maitiro, senge a medallion architecture uye yakasimba semantic layer. Nekudaro, isu taisawanzowana masangano aitoita izvi muhupenyu chaihwo. Mazhinji masangano anoshandisa maspredishiti, hafu-yakabikwa matafura, uye inogara ichichinja data data. Kubva pano, takauya nepfungwa yekuvaka nyanzvi dzeAI vamiririri vanogona kuvakwa nekukurumidza kuti vapindure seti yemibvunzo.

Sezvo makambani anokura, anobata data rakawanda uye ane vashandisi vakawanda. Iyo agent mesh pfungwa inobatsira kuenzanisa nekukurumidza kuita sarudzo nekutonga kunodiwa mukutonga. Specialized agents vanobatsira kuseta miganhu yakajeka uye mabasa kune yega yega AI. Ivo zvakare vanogadzira scalable nzira yekuti vamiririri vataure. Uyezve, ivo vanogona kubatsira kubata zviwanikwa zvakanaka muzvikwata nemakambani.

Specialized AI vamiririri

Pfungwa iri kuseri kweanyanzvi nderekuti mumiriri uyu anogona uye anongopindura mibvunzo pane dataset rakanyatsotsanangurwa. Semuenzaniso, unogona kugadzira uye kuvhura mumiriri weAI anopindura mibvunzo nezve mishandirapamwe yekushambadzira. Kana iwe unogona kuvaka imwe yekupindura mibvunzo nezve pombi yekushambadzira, zvichingodaro uye zvichingodaro.

Isu munguva pfupi yapfuura takatanga Agent Analyst , uchishandisa chivakwa ichi. Zviratidzo zvekutanga zvinovimbisa zvikuru. Kana madataset akanyatso kucheneswa uye padanho rekurudyi, vamiririri ava vanogona kupindura seti yemibvunzo yakavimbika zvakanyanya. Vavaki veava vamiririri vanogona kugovana navo nevasiri-tekinoroji vashandisi uye kuzorora zviri nyore kuziva kuti iyo AI haipindure mibvunzo iri kunze kwechikamu.

Pane chikanganiso chimwe chete: Vashandisi vanofanirwa kuziva kuti ndeupi mumiririri wekuenda kumubvunzo upi. Zvakafanana nekuda kuziva muongorori wekutengesa kuti abvunze mubvunzo we vs. kungobvunza mubvunzo wakajairika. Nemubvunzo wakajairika, mumwe munhu ari muchikwata anogona kuitungamira kune munhu akakodzera. Apa ndipo panopinda pfungwa ye "agent mesh".

Kubatanidza vamiririri pamwe chete

Kana mumiririri mumwe chete achigona kupindura akavimbika mibvunzo yakanangana nedomasi, saka wadii kurega vamiririri vachitaurirana? Sei usingakwanise, semuenzaniso, mumiriri wekushambadzira anongobvunza mumiriri wepombi zvakananga kana vachigona kupindura mubvunzo zviri nyore? Tinotenda kuti inofanira kukwanisa. Sezvineiwo, isu tinofunga kuti mune ramangwana kuchava nema network evamiririri vane hierarchical structure. Unogona kufungidzira "GTM agent" inodana "Marketing agent." Mumiririri uyu anobva adaidza vese "Pipeline agent" uye "Marketing campaign agent."

Pfungwa iyi yakafanana nepfungwa yakajairika inotenderera ichitenderedza AI inozivikanwa se " Indaneti yeAgents ." Iramangwana apo vamiririri veAI vanobatana zvakanaka mumasangano akasiyana siyana. Vanoita izvi vachiona kuti kuchengetedzeka uye kuvimbana kunoramba kwakasimba.

Iyi mesh nzira inopa mashoma akakosha mabhenefiti pamusoro pe monolithic AI (pane pristine semantic layer):

• Kucherechedzwa: Sezvo mumiririri mumwe chete achipa mhinduro zvichienderana nedata chairo, unogona kutsvaga mhinduro yega yega kudzokera kumumiririri iyeye. Izvi zvinobatsira kuve nechokwadi chechokwadi kuburikidza nekuongorora. Kupa kongiri, kunyangwe zvakapusa zvakanyanya, semuenzaniso, fungidzira uine matafura ezviitiko zviviri: imwe yekushambadzira uye imwe yechigadzirwa. Kana mushandisi akabvunza, "Ndezvipi zviitiko zvakaburitsa mari yakawanda?" iyo AI inogona kufunga kuti inoreva zviitiko zvechigadzirwa. Kunyangwe zvisizvo, mushandisi anogona kuona kuti ndeupi mumiririri akapindura uye anogona kutungamira iyo AI.
• Kuchengetedza: Kungofanana neinjini yemotokari, kana iwe uchigona kuwana matambudziko nyore uye nekukurumidza kutsiva zvikamu, mota inova yakavimbika. Kana mumiriri mumwe akatanga kutadza nekuda kwekuchinja kweiyo data modhi, zvino inogona kukurumidza kuoneka uye iyo mumiriri inogona kuvandudzwa.
• Kurongeka: Nemumiriri wega wega uchishanda mukati mezvivharo zvayo, hapana nzvimbo yekuti ibve panjanji. Iyo inogona kunge isina mhinduro, asi haizoite chimwe chinhu chekufungidzira.

Pakupera kwezuva, iyi pfungwa ye mesh haisi novel. Izvi zvinoratidzira pfungwa yemusanganiswa wenyanzvi uyo wakaratidzwa kuvandudza humbowo hweLLMs. Iko kungotora iyo pfungwa imwe chete uye kuiunza kune AI vamiririri.

Zvinetso zvehunyanzvi zveagent meshes

PaFabi.ai, tine rwendo rurefu rwekufamba sezvatinovaka Analyst Agent mesh. Asi, isu takatokunda mamwe ematambudziko makuru ehunyanzvi hwekuvaka munzira.

AI data analyst agents vanoda yakasarudzika dhizaini. Iyi dhizaini inofanirwa kuvabvumira kushandisa Python kana LLMs kupindura mibvunzo, kugara mukuwirirana nedata masosi, uye kukwana mumapuratifomu anodyidzana, vachiri kugara vakachengeteka uye scalable. Mumwe nemumwe mumiririri anofanirwa kushanda muPython kernel yake, iyo inoda kukurumidza kukwizwa kumusoro kana pasi kuti ideredze mitengo uye irambe yakabatana neyekunobva data.

Mavakirwo asingapi kernels ega kune mumiririri wega wega anogona kupinda mune imwe yeinotevera njodzi:

• State kunetsana kwezvakasiyana pakati peAI vamiririri. Vamiririri vaviri vakasiyana vanogona kugadzira "foo" mutsauko kupindura mubvunzo, zvichikonzera kusawirirana. Panogona kuve nedzimwe nzira dzekugadzira zviziviso zvakasiyana, asi izvi zvinowedzera kusawirirana kweAI kugadzira kodhi isiriyo.
• Njodzi dzekuchengetedza dzinokonzerwa nekugovana data pakati pezvikwata zvakasiyana kana kunyange masangano akasiyana.
• Kuita zvinokanganisa kana mumiririri atora huwandu husina kuenzana hwemakomputa zviwanikwa.

Dambudziko rekuvaka rudzi urwu rwepuratifomu idambudziko reAI sezvo iri DevOps dambudziko.

Kutarisa kumberi: Kumbundira nyanzvi, inotongwa AI vamiririri mune data

Sezvo makambani emabhizinesi achitarisira mamwe maAI maapplication mukushanda kwavo, vanoda nzira dzakanyanya uye dzakanyatsotongwa. Iyo mumiriri mesh chimiro inoshandisa nyanzvi yeAI data vamiririri senzira yekuyera AI mune data analytics. Iyi nzira inochengetedza chengetedzo, kuvimbika, uye kuita kwakasimba.

Tinogona kunge takatarisira kuti AI ichave kwese kwese ikozvino, ichipindura mibvunzo yakawanda yedata. Asi, kana tikanyatsotarisisa, kufambira mberi mumakore maviri chete kubva pakatangwa ChatGPT kunoshamisa. Tichine zvakawanda zvekudzidza murwendo urwu. Mupfungwa dzangu, zvisinei, vamiririri uye agent mesh masisitimu anozove akakosha kune bhizinesi AI.