[{"data":1,"prerenderedAt":334},["ShallowReactive",2],{"project-40998497":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":28,"sentiment":29,"stance_mix":30,"scored_comments":34,"aspects":35,"strip":43,"post_text":45,"what":46,"said":70,"maker_in_thread":132,"shape":144,"grid":149,"attempts":297,"same_week":298,"maker_shows":194,"maker_hidden":326,"reader_version":327,"card":328},"40998497","Mutable.ai Auto Wiki","How we leapfrogged traditional vector based RAG with a 'language map'","title","Show HN: How we leapfrogged traditional vector based RAG with a 'language map'","https:\u002F\u002Ftwitter.com\u002Fmutableai\u002Fstatus\u002F1813815706783490055","twitter.com","oshams",1721328014,2024,162,55,507,3515,15,"notable","ai","code assistant","web_app",[24,25,26,27],"codebase chat","rag","code wiki","llm code assistant","mixed",0.154,{"positive":31,"neutral":31,"mixed":32,"critical":33},7,1,3,18,{"idea":36,"design":37,"performance":39,"execution":40,"pricing":41},{"positive":33,"critical":32},{"positive":38,"critical":38},0,{"positive":32,"critical":38},{"positive":32,"critical":38},{"positive":38,"critical":42},2,[32,38,38,38,38,38,38,38,38,38,33,38,44,38,38,38,38,38,38,44,44,33,33,38,38,32,38,32,38,33,33,33,38,33,38,44,44,38,44,44],5,"TL;DR: Vector-based RAG performs poorly for many real-world applications like codebase chats, and you should consider 'language maps'.\n\nPart of our mission at Mutable.ai is to make it much easier for developers to build and understand software. One of the natural ways to do this is to create a codebase chat, that answer questions about your repo and help you build features.\n\nIt might seem simple to plug in your codebase into a state-of-the-art LLM, but LLMs have two limitations that make human-level assistance with code difficult:\n\n1. They currently have context windows that are too small to a",{"problem":47,"audience":53,"pricing":57,"motivation":61,"time_to_build":65,"stack":66,"openness":67,"stage":68,"team_size":69},{"value":48,"source":49,"evidence":50},"Vector-based RAG gives poor answers for codebase questions","maker",[51],{"id":52},"post",{"value":54,"source":49,"evidence":55},"developers asking questions about their code repositories",[56],{"id":52},{"value":58,"source":49,"evidence":59},"Free for open source repos; starts at $2\u002Fmo per private repo",[60],{"id":52},{"value":62,"source":49,"evidence":63},"Make it much easier for developers to build and understand software",[64],{"id":52},null,[],"unknown","launched","company",{"praise":71,"objections":88,"requests":107,"questions":112,"alternatives":123},[72],{"value":73,"source":74,"evidence":75},"Commenters found the wiki approach interesting and effective","commenter",[76,80,84],{"id":77,"by":78,"text":79},"41003540","senko","Looks similar to what we're doing in Pythagora with CodeMonkey agent (prompt: https:\u002F\u002Fgithub.com\u002FPythagora-io\u002Fgpt-pilot\u002Fblob\u002Fmain\u002Fcore\u002Fprompts\u002Fcode-monkey\u002Fdescribe_file.prompt,…",{"id":81,"by":82,"text":83},"41004504","kleneway1","Nice job on this, it’s a really interesting approach. I’ve been developing an open-source coding agent over the past year, and RAG just wasn’t working at all. I switched to a…",{"id":85,"by":86,"text":87},"41007586","J_Shelby_J","This is amazing. Do you have any information on the implementation?",[89,100],{"value":90,"source":74,"evidence":91},"Login required to ask questions",[92,96],{"id":93,"by":94,"text":95},"41001803","spirobelv2","it wants me to login to ask a question I will just keep using phind you have vc dollars - sponsor a public free search over open source repos. Also think about what happens when…",{"id":97,"by":98,"text":99},"41010153","fudged71","This comes across as pretty rude... they're already indexing and generating pages for public repos for free, and providing this search for free, just behind a login. If you want…",{"value":101,"source":74,"evidence":102},"Wiki may be too wordy and not token efficient",[103],{"id":104,"by":105,"text":106},"41002519","anotherpaulg","I agree that many AI coding tools have rushed to adopt naive RAG on code. Have you done any quantitative evaluation of your wiki style code summaries? My first impression is that…",[108],{"value":109,"source":74,"evidence":110},"Open access without login for public repos",[111],{"id":93,"by":94,"text":95},[113],{"value":114,"source":74,"evidence":115},"How the wiki is generated and structured",[116,120],{"id":117,"by":118,"text":119},"41009421","billmalarky","Can you share some basic insight into how your system processes an open source repo to generate a wiki with a hierarchy and structure that maps to the same or similar hierarchy…",{"id":121,"by":118,"text":122},"41009743","Do you build a syntax tree of the code, then loop the tree to auto-write an article for each node (or the larger or more material nodes) and then also reference the tree to pull…",[124,128],{"value":125,"source":74,"evidence":126},"Aider repo map",[127],{"id":104,"by":105,"text":106},{"value":129,"source":74,"evidence":130},"Phind",[131],{"id":93,"by":94,"text":95},{"comments":133,"answered":31,"key":134},14,[135,138,141],{"id":136,"text":137},"41007256","Nice!",{"id":139,"text":140},"41007230","That's awesome thanks!",{"id":142,"text":143},"41007247","patience ... all part of our plan :)",{"commenters":145,"max_depth":146,"maker_replies":133,"first_comment_mins":147,"lifespan_hours":148},28,6,525,60,[150,152,156,158,161,164,168,172,174,177,180,182,185,186,190,192,196,200,201,204,207,210,214,216,219,221,224,227,230,235,238,240,242,246,249,252,255,256,258,262,266,269,271,274,277,279,282,283,286,288,290,293,295],[104,38,32,151,105,38],"project\u002Fcritical\u002Fidea",[153,38,38,154,155,38],"41002895","commenter\u002Fneutral\u002Fidea","lemming",[157,38,38,154,105,38],"41003014",[159,38,38,160,155,38],"41003247","commenter\u002Fneutral\u002Fnone",[162,38,38,160,163,38],"41005637","emporas",[165,38,38,166,167,38],"41007018","commenter\u002Fcritical\u002Fexecution","manishsharan",[169,38,38,170,171,38],"41007169","other\u002Fneutral\u002Fnone","danenania",[173,38,38,160,105,38],"41009262",[175,38,38,170,176,38],"41014410","paradite",[178,32,38,154,179,38],"41005459","lmeyerov",[136,32,38,181,11,32],"maker\u002Fpositive\u002Fnone",[183,42,38,170,184,38],"41006901","kordlessagain",[139,42,38,181,11,32],[187,33,33,188,189,38],"41002509","project\u002Fneutral\u002Fdesign","tlarkworthy",[142,33,38,191,11,32],"maker\u002Fneutral\u002Fnone",[193,194,44,181,195,38],"41008857",4,"maxsieg",[197,194,38,198,199,38],"41014885","commenter\u002Fpositive\u002Fidea","abdullin",[77,44,38,198,78,38],[202,146,38,154,203,38],"41002186","langcss",[205,146,38,154,206,38],"41002613","PhilippGille",[208,146,38,170,209,38],"41003119","zbyforgotp",[211,146,38,212,213,38],"41003434","commenter\u002Fcritical\u002Fnone","janalsncm",[215,146,38,160,209,38],"41023037",[217,146,38,154,218,38],"41010797","koliber",[220,146,38,181,11,32],"41007341",[222,31,44,223,195,38],"41007804","project\u002Fpositive\u002Fperformance",[81,225,44,226,82,38],8,"project\u002Fpositive\u002Fidea",[228,229,38,191,11,32],"41007176",9,[231,232,33,233,234,38],"41009526",10,"project\u002Fmixed\u002Fexecution","zellyn",[236,232,33,237,234,38],"41009664","project\u002Fneutral\u002Fnone",[239,232,38,181,11,32],"41010058",[241,232,38,237,11,32],"41011220",[243,244,38,170,245,38],"41010607",11,"zby",[93,247,32,248,94,38],12,"project\u002Fcritical\u002Fpricing",[250,247,38,251,11,32],"41007292","maker\u002Fneutral\u002Fpricing",[253,247,32,254,94,38],"41022384","maker\u002Fcritical\u002Fpricing",[97,247,38,160,98,38],[257,247,38,160,94,38],"41022358",[259,260,33,237,261,38],"41005645",13,"magicalhippo",[263,260,33,264,265,38],"41006264","project\u002Fneutral\u002Fidea","songeater",[267,260,33,264,268,38],"41006720","kaycebasques",[270,260,33,264,261,38],"41006868",[272,260,38,273,11,32],"41007276","maker\u002Fneutral\u002Fdesign",[275,260,33,276,261,38],"41010341","project\u002Fneutral\u002Fpricing",[278,260,38,160,11,32],"41007416",[280,133,38,281,11,32],"41007796","maker\u002Fpositive\u002Fpricing",[117,133,44,226,118,38],[284,133,38,285,11,32],"41009689","maker\u002Fpositive\u002Fexecution",[121,133,44,287,118,38],"project\u002Fpositive\u002Fexecution",[289,133,38,191,11,32],"41009792",[291,133,44,292,118,38],"41009963","maker\u002Fpositive\u002Fidea",[85,18,44,294,86,38],"project\u002Fpositive\u002Fnone",[296,18,38,191,11,32],"41009098",[],[299,304,309,313,317,321],{"id":300,"name":301,"points":302,"category":303},"41007403","Sendune",429,"web_saas",{"id":305,"name":306,"points":307,"category":308},"41045834","Briefer",265,"data",{"id":310,"name":311,"points":312,"category":303},"41031975","billabear.com",260,{"id":314,"name":315,"points":316,"category":20},"41048194","Zerox",246,{"id":318,"name":319,"points":320,"category":303},"41048196","Overlap",229,{"id":322,"name":323,"points":324,"category":325},"41024964","Moocable",161,"consumer_social",false,"v2",{"id":329,"name":5,"line":6,"ai":326,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":330,"top":18,"kw":331,"maker":11,"domain":10,"tier":19,"attempts":38,"strip":332,"url":9,"mix":333,"praise":73,"objection":90,"sub":21},40998497,0.15,[24,25,26,27],"1000000000305000000553300101033303055055",[31,31,32,33],1794036658045]