hakia

» see hakia.com as it lived, on the wayback machine

~ the eulogy ~

the first time i heard of hakia was around 2007 on turkish tech forums. someone posted "türk kurucu, semantic search, google'a rakip" and the frmtr thread had the usual mix of pride and skepticism. rıza berkan is a nuclear physicist by training with a phd in ai and fuzzy logic. he founded hakia in march 2004 out of new york with pentti kouri, a finnish economist and vc. the pitch was that keyword search matches strings, semantic search understands meaning, and if you built the ontology carefully enough you could beat google on the questions that mattered most.

the tech was called qdexing, with a rank algorithm called semanticrank on top. under the hood it was ontological semantics, fuzzy logic, some computational linguistics, and a fair amount of hand-tuned math, running against a curated index of pages that hakia scored for credibility as well as relevance. the demo case was medical queries: type "why do i get migraines after coffee" and get answer-style results from vetted health sites, not a wikipedia link plus ten forum threads. the 2008 relaunch added a "credible sites" toggle, image search, and a "meet others" tool for finding forums on the same topic. tech press paid attention. the search engine journal wrote it up. arnoldit did a founder interview.

none of it converted into traffic. google was untouchable for general search by 2008 and any startup that wanted to unseat it needed either a wedge category that google could not match, or enough capital to survive a decade of quiet iteration. hakia had ideas but not that wedge. bing entered the same year with microsoft money behind it and still lost. the vertical pivot to medical/legal/education segments came in 2010 and did not produce recurring revenue either. the site went dormant around 2014. the domain got sold to hughesnet the year after. the current hakia.com is a satellite internet marketing shell with nothing left of the search engine attached.

~ the dossier ~

~ the turkish angle ~

rıza can berkan was born in turkey, did his physics phd at michigan, and moved into ai research at brookhaven. hakia had a turkish co-founder and a new york office and a lot of turkish engineers on the roster in the early years. for someone growing up on ttnet dial-up and reading donanimhaber, "türk mühendis google'a kafa tutuyor" was the kind of headline you clicked. it was one of the few times a turkish-founded american startup made an actual dent in mainstream us tech coverage before the acqui-hire era.

~ ten years too early ~

the honest read on hakia is that its bet was correct and its timing was wrong. semantic search as a category is now the default: google's featured snippets, wolfram alpha, bing chat, perplexity, chatgpt browsing all do what hakia's 2008 demo did, better, using tools that did not exist yet in the mid-2000s. hakia was building ontologies by hand when the industry was about to shift to learned embeddings. by the time gpt-1 came out in 2018 there was no one left at hakia to notice.

~ what it left ~

berkan has resurfaced periodically talking about the history of semantic search on a subdomain of the hakia name, and the qdex patents got cited in later work on structured retrieval. there is no product to point to. the company was worth ~$4m at its 2007 series a and never got past a series b conversation. the ten-year tail is the story. it is one of the longest-running dead-in-place search engines from the pre-neural era.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »