{"id":367,"date":"2025-01-31T22:56:00","date_gmt":"2025-01-31T22:56:00","guid":{"rendered":"http:\/\/blog.firatyasar.com\/?p=367"},"modified":"2025-03-29T22:59:47","modified_gmt":"2025-03-29T22:59:47","slug":"agentic-ai-data-stores","status":"publish","type":"post","link":"https:\/\/blog.firatyasar.com\/?p=367","title":{"rendered":"Agentic AI &#8211; Data Stores"},"content":{"rendered":"\n<p>Bir dil modelini dev bir k\u00fct\u00fcphane gibi hayal edin. \u0130\u00e7inde e\u011fitildi\u011fi verilerden olu\u015fan milyonlarca sayfa bilgi bulunur. Ancak bu k\u00fct\u00fcphane zamanla g\u00fcncellenmez; sadece ba\u015flang\u0131\u00e7ta sahip oldu\u011fu bilgiyle s\u0131n\u0131rl\u0131d\u0131r. Oysa ger\u00e7ek d\u00fcnya s\u00fcrekli de\u011fi\u015fmektedir. Yeni bilgiler, geli\u015fen olaylar ve kullan\u0131c\u0131 ihtiya\u00e7lar\u0131 modele daha dinamik ve g\u00fcncel veri kaynaklar\u0131na eri\u015fim gereksinimini do\u011furur. \u0130\u015fte bu noktada devreye <strong>Data Stores<\/strong> girer.<\/p>\n\n\n\n<h3>Dinamik Bilgiye A\u00e7\u0131lan Kap\u0131<\/h3>\n\n\n\n<p>\u00d6rne\u011fin bir geli\u015ftirici, modele destekleyici bilgi olarak bir Excel tablosu, bir PDF ya da \u015firket i\u00e7i bir belge sunmak isteyebilir. Bu gibi durumlarda Data Store yap\u0131s\u0131 sayesinde:<\/p>\n\n\n\n<ul><li>Veriler orijinal format\u0131nda sunulabilir.<\/li><li>Zaman al\u0131c\u0131 veri d\u00f6n\u00fc\u015ft\u00fcrme i\u015flemleri gerekmez.<\/li><li>Modeli yeniden e\u011fitmeye veya fine-tuning yapmaya ihtiya\u00e7 kalmaz.<\/li><\/ul>\n\n\n\n<p>Data Store, bu belgeleri otomatik olarak <strong>vector embeddings<\/strong> \u015feklinde vekt\u00f6r veritaban\u0131na d\u00f6n\u00fc\u015ft\u00fcr\u00fcr. Bu yap\u0131 sayesinde agent, bu g\u00f6m\u00fclere dayal\u0131 olarak kullan\u0131c\u0131 sorgusuyla ili\u015fkili en do\u011fru bilgiyi bulabilir ve bir sonraki eylemini veya cevab\u0131n\u0131 bu bilgiyle destekleyebilir.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" width=\"1024\" height=\"418\" src=\"\/wp-content\/uploads\/2025\/03\/image-20-1024x418.png\" alt=\"\" class=\"wp-image-368\" srcset=\"\/wp-content\/uploads\/2025\/03\/image-20-1024x418.png 1024w, \/wp-content\/uploads\/2025\/03\/image-20-300x122.png 300w, \/wp-content\/uploads\/2025\/03\/image-20-768x314.png 768w, \/wp-content\/uploads\/2025\/03\/image-20-1536x627.png 1536w, \/wp-content\/uploads\/2025\/03\/image-20-660x269.png 660w, \/wp-content\/uploads\/2025\/03\/image-20.png 1788w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<h3>Data Store\u2019un Mimarideki Yeri<\/h3>\n\n\n\n<p>Generative AI agent\u2019lar\u0131 ba\u011flam\u0131nda, <strong>Data Store<\/strong>, genellikle \u00e7al\u0131\u015fma zaman\u0131nda eri\u015filen bir <strong>vekt\u00f6r veritaban\u0131<\/strong> olarak i\u015flev g\u00f6r\u00fcr. Bu veritabanlar\u0131, metinleri say\u0131sal vekt\u00f6r temsillerine d\u00f6n\u00fc\u015ft\u00fcrerek bilgiye daha semantik bir d\u00fczeyde eri\u015fim sa\u011flar.<\/p>\n\n\n\n<p>Bu yakla\u015f\u0131m \u00f6zellikle <strong>Retrieval Augmented Generation (RAG)<\/strong> tabanl\u0131 sistemlerde yayg\u0131n olarak kullan\u0131l\u0131r. RAG yap\u0131s\u0131, dil modelinin yaln\u0131zca e\u011fitildi\u011fi bilgiye de\u011fil, g\u00fcncel ve g\u00f6revle ilgili verilere eri\u015fmesini sa\u011flar.<\/p>\n\n\n\n<p>Veriler a\u015fa\u011f\u0131daki gibi \u00e7e\u015fitli kaynaklardan gelebilir:<\/p>\n\n\n\n<ul><li>Web sitesi i\u00e7erikleri<\/li><li>Yap\u0131land\u0131r\u0131lm\u0131\u015f veriler (PDF, Word, Excel, CSV vs.)<\/li><li>Yap\u0131land\u0131r\u0131lmam\u0131\u015f metinler (HTML, d\u00fcz TXT dosyalar\u0131 vs.)<\/li><\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" width=\"1024\" height=\"212\" src=\"\/wp-content\/uploads\/2025\/03\/image-21-1024x212.png\" alt=\"\" class=\"wp-image-369\" srcset=\"\/wp-content\/uploads\/2025\/03\/image-21-1024x212.png 1024w, \/wp-content\/uploads\/2025\/03\/image-21-300x62.png 300w, \/wp-content\/uploads\/2025\/03\/image-21-768x159.png 768w, \/wp-content\/uploads\/2025\/03\/image-21-1536x319.png 1536w, \/wp-content\/uploads\/2025\/03\/image-21-2048x425.png 2048w, \/wp-content\/uploads\/2025\/03\/image-21-660x137.png 660w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<h3>Bir Data Store Sorgusunun Tipik Ak\u0131\u015f\u0131<\/h3>\n\n\n\n<p>A\u015fa\u011f\u0131da bir kullan\u0131c\u0131 sorgusunun Data Store arac\u0131l\u0131\u011f\u0131yla nas\u0131l i\u015flendi\u011fini ad\u0131m ad\u0131m g\u00f6rebilirsin:<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" width=\"1024\" height=\"636\" src=\"\/wp-content\/uploads\/2025\/03\/image-22-1024x636.png\" alt=\"\" class=\"wp-image-370\" srcset=\"\/wp-content\/uploads\/2025\/03\/image-22-1024x636.png 1024w, \/wp-content\/uploads\/2025\/03\/image-22-300x186.png 300w, \/wp-content\/uploads\/2025\/03\/image-22-768x477.png 768w, \/wp-content\/uploads\/2025\/03\/image-22-660x410.png 660w, \/wp-content\/uploads\/2025\/03\/image-22.png 1082w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ol><li>Kullan\u0131c\u0131 sorgusu, embedding modeli arac\u0131l\u0131\u011f\u0131yla g\u00f6m\u00fcye (embedding) d\u00f6n\u00fc\u015ft\u00fcr\u00fcl\u00fcr.<\/li><li>Bu embedding, <strong>SCaNN<\/strong> gibi bir e\u015fle\u015ftirme algoritmas\u0131 kullan\u0131larak vekt\u00f6r veritaban\u0131ndaki i\u00e7eriklerle kar\u015f\u0131la\u015ft\u0131r\u0131l\u0131r.<\/li><li>E\u015fle\u015fen i\u00e7erik, metin format\u0131nda vekt\u00f6r veritaban\u0131ndan geri al\u0131n\u0131r.<\/li><li>Agent, hem kullan\u0131c\u0131 sorgusunu hem de e\u015fle\u015fen i\u00e7eri\u011fi al\u0131r ve bir yan\u0131t veya aksiyon \u00fcretir.<\/li><li>Son olarak kullan\u0131c\u0131ya nihai cevap iletilir.<\/li><\/ol>\n\n\n\n<h3>Etkin ve G\u00fcvenilir Yan\u0131tlar \u0130\u00e7in<\/h3>\n\n\n\n<p>Bu yap\u0131 sayesinde, agent\u2019lar kullan\u0131c\u0131n\u0131n sorgusunu bilinen bir bilgi kayna\u011f\u0131 ile e\u015fle\u015ftirerek, sabit e\u011fitim verilerine tak\u0131l\u0131 kalmadan, daha do\u011fru, ba\u011flama uygun ve g\u00fcncel yan\u0131tlar \u00fcretebilirler. Gerekirse bu d\u00f6ng\u00fc tekrarlanabilir ve agent, ikinci bir arama yaparak sonucu daha da rafine edebilir.<\/p>\n\n\n\n<h2>Sonu\u00e7<\/h2>\n\n\n\n<p>Data Stores, dil modellerinin dura\u011fan yap\u0131s\u0131n\u0131 dinamik ve ba\u011flama duyarl\u0131 bir bilgi eri\u015fim altyap\u0131s\u0131na d\u00f6n\u00fc\u015ft\u00fcren kritik bir bile\u015fendir. Agent mimarisinde bu yap\u0131, kullan\u0131c\u0131n\u0131n ger\u00e7ek zamanl\u0131 sorgular\u0131na g\u00fc\u00e7l\u00fc ve kan\u0131ta dayal\u0131 yan\u0131tlar \u00fcretme imk\u00e2n\u0131 sunar. \u00d6zellikle dok\u00fcman a\u011f\u0131rl\u0131kl\u0131 sekt\u00f6rlerde (finans, hukuk, teknik destek, vb.) Data Store kullan\u0131m\u0131, model performans\u0131n\u0131 ciddi oranda iyile\u015ftirme potansiyeline sahiptir.<\/p>\n\n\n\n<p><strong>Kaynak<\/strong>: AI Agent Books \u2013 Authors: Julia Wiesinger, Patrick Marlow and Vladimir Vuskovic<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Bir dil modelini dev bir k\u00fct\u00fcphane gibi hayal edin. \u0130\u00e7inde e\u011fitildi\u011fi verilerden olu\u015fan milyonlarca sayfa bilgi bulunur. Ancak bu k\u00fct\u00fcphane zamanla g\u00fcncellenmez; sadece ba\u015flang\u0131\u00e7ta sahip oldu\u011fu bilgiyle s\u0131n\u0131rl\u0131d\u0131r. Oysa ger\u00e7ek d\u00fcnya s\u00fcrekli de\u011fi\u015fmektedir. Yeni bilgiler, geli\u015fen olaylar ve kullan\u0131c\u0131 ihtiya\u00e7lar\u0131 modele daha dinamik ve g\u00fcncel veri kaynaklar\u0131na eri\u015fim gereksinimini do\u011furur. \u0130\u015fte bu noktada devreye Data\u2026 <span class=\"read-more\"><a href=\"https:\/\/blog.firatyasar.com\/?p=367\">Read More &raquo;<\/a><\/span><\/p>\n","protected":false},"author":1,"featured_media":348,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[149,159,85,150,133],"_links":{"self":[{"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/posts\/367"}],"collection":[{"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=367"}],"version-history":[{"count":2,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/posts\/367\/revisions"}],"predecessor-version":[{"id":372,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/posts\/367\/revisions\/372"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=\/wp\/v2\/media\/348"}],"wp:attachment":[{"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=367"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=367"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.firatyasar.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=367"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}