{
  "video_id": "b2RInLZB9ms",
  "channel_slug": "assemblyai",
  "channel_handle": "assemblyai",
  "title": "Building Needle: Vibe automation, RAG workflows, and founder lessons",
  "duration_seconds": 2165.0,
  "url": "https://www.youtube.com/watch?v=b2RInLZB9ms",
  "upload_date": "",
  "transcript": "Hi everyone, my name is Mart. I'm from\nAssembly and I'm joined here by Yan from\nNeato. Um, today we'll just be talking\nabout uh startups building with voice AI\nand Yen's experience as well um at\nNeato. Yan, thank you so much for\njoining me today. Can you mind do you\nmind introducing yourself for everyone\nhere?\nYeah, also happy to be here and talking\nwith you, man. I'm John uh from Needle\num and uh we are building automations.\nSo we want to make it possible that\neverybody can build automations. So we\nare basically a vibe automation platform\nsuch as you know vibe coding is quite of\na thing to allow everybody to build like\ncool applications with code and lovable\nis maybe one of the leading uh products\nin this case or cursor. Um I think\nthere's a massive new emerging market\nwhich is like automations. I think 2026\nis the year of automations and agents.\nBut to build these sometimes you may\nhave to use a Python SDK or you use some\nVersel SDK or maybe you need to be\nrelatively technical configuring JSON\nobjects and the people who really want\nto build automations are or know how\nwhich automations to build are not\nnecessarily the technical ones but the\nones who actually are in the field and\nhence we want to give them a tool to be\nable to build automations for their use\ncases and so yeah that's why we created\nNeedle\num a VIP automation platform so with\nbuilt-in rack and we also have an\nintegration with assembly AI um which is\nactually also we do some hackathons so\nwe started this 3 months ago and since\nthen we're doing hackathons and we have\naround 50 participants normally in the\nhackathons and we also see that assembly\na is a relatively popular uh integration\nin the in the hackathons which people\nlike to use in specific case together\nwith telegram because let's say you want\nto send a voice note on telegram and You\nwant to trigger use this as a trigger\nand then you may want to use assembly A\nto convert this voice to text and give\nit to an AI agent because AI agent\ncannot work with audio has to work with\ntext and then maybe do like a specific\nsearch across your Google drive files to\nfind an answer for the question that\nperson asked on telegram via voice and\nthen may want to convert this back uh to\nvoice uh um with\nassembly AI and then send back a message\nin telegram. So I see this for example\nas a classic use case with the seminary\nthat's why I reached out and I thought\nmaybe could demo this kind of use case\ntogether because I think it's also very\nrelevant for like let's say customer\nsupport on WhatsApp, customer support on\ntelegram or so and that's kind of like\nwhy I wanted to chat with you and so\nmaybe we can see if there's some\npotential of collab or something like\nthis.\nYeah, great. Thanks for the great intro.\nI um I wanted to say it kind of reminds\nme of like I guess the tools that we use\ninternally kind of like um N8N and\nZapia. Um but the difference kind of\nlooks like uh there's a rag built in. So\nyou're doing like an active search over\nuh a knowledge base.\nCan you maybe kind of tell me a bit more\nabout that? Um I'm kind of used to\nworking with Zapia or NN.\nYeah, sure. I mean there's two\ndifferentiation factors. So the first\none is that as you said rack we started\nbasically as a rack API and became a\nrack for builder. So our initial product\nwas just rack API only because uh in\n2025 so last year we saw many people\nwant to build rack applications and it's\nrelatively complex maybe because let's\nsay you have to set up a vector database\nyou have to think of your chunking\nstrategy let's say your data lives in\nthe email. Oh, you have to do a lot of\nwork to in the end you don't care. You\n[ __ ] just want to share a customer\nsupport chatbot. You don't want to spend\nlike uh two weeks on setting up and\nmaintaining also maintaining also the\nsoftware uh for that. So that's why we\nwere like starting with the rack API and\nwe realized many people want to do\nrepetitive tasks with the rack API which\nis for example answer the email you know\nso based on my knowledge base uh and\ninstead of them manually going ahead and\ndoing a semantic search to find the\nrelevant ticket for example for this\nemail uh we we thought okay it seems to\nbe like there's a huge demand for\nbuilding for this automating with rack\nand so that's why we built then the\nautomation platform And um yeah, so if\nyou have ever set up a rack kind of\nworkflow with other uh automation tools,\nmaybe N8N uh I think N8N is a great\ntool. I'm I'm not like thinking is a bad\ntool. It's super cool and also has a\ngreat community. Uh and um we're in a\nsimilar space, right? But if you have\never set for example rack and niten, I\nthink you need like I don't know eight\nnodes or something. You need to you have\nsome configuration issues with like the\nmaybe use quadrat vector database or pg\nvector and then you have to think\nyourself of like actually all this\ntechnical stuff and if you actually just\nwant to quickly and ship a high quality\nalso solution that for example in the\nhackathon for example we had like with\nthe you just want to have this telegram\nsupport and have this search kind of\nthrough your knowledge base which is\nmaybe your zenes tickets which you have\njust automatically indexed with needle\nand you can do record them or maybe your\num Jira tickets or confluence documents\num then you you don't want to set all\nthis up and you anyway would never be\nable to do in such a short time and so\nthe overhead of setting all this up is\ntoo big that some people just let go and\nI mean you maybe follow also Reddit and\nso many people on Reddit I actually say\noh I tried to set up this like rag\npipeline in it and I have some problem\ncan somebody help me out it's like\nmassive amount amount of people asking\nthis in different channels in the Nitn\nchannel or the rack channel or the\nclassic uh Reddit channels and I think\nit's one of the interesting sources also\nto know about what people struggle with\nand so yeah the needle is super easy you\njust have like one note which is needle\nnote the needle agent which has access\nto then [snorts] a collection and a\ncollection is a set of files that you\nconnected let's say to different sources\nGoogle drive notion uh confluence and\nmaybe slack depends on what you want to\nconnect you decide\nAnd then you say okay agent you have\naccess to this and we do have indexed\nthe data. So we do rack on this. You can\nnow use the rack tool do a deep search\non this based on the input that you get\nand then give me back an answer and\nthat's it. So it's just one note instead\nof like six seven eight notes and you\ndon't have to configure anything. You\njust have to click click click and then\nit's ready to go. So it's we often work\ntogether with like uh\nmedium-sized teams um consultants um\nmarketing agencies which are somewhat\ntechnical. They are liking to use\ntechnical solutions but they are not\nnecessarily product uh software\nengineers. So in that case that is kind\nof like the suiting solution for them.\nYeah voice is right now a big market. So\nyou're doing you're in a great market\nassembly AI I mean we are not like\nnecessarily voice we are just like kind\nof like uh helping to stick together\nsolutions right like your solution is\none of the parts of the total solution I\nmean Google drive for example would be\nstill needed right but um I think it's\nan exciting market you are in and very\ncool also recently very hot uh miss 11\nlaps also I think they are they went\npretty big uh right now they just\nannounced another round I two days ago\nor something right 500 mil at 11 bit\nevaluation or something like this. So\nvoice exciting market I think also voice\nagents voice customer support. So I\nthink I will see more uh and more of\nthese kind of workflows also built by\nusers within needle. Um yeah in this\nthere's a fair amount of overlap like um\nwe get lots of users building voice\nagents who need to do rag over like you\nknow a series of customer service SOPs\nor u maybe they have a knowledge base.\nWe we internally have also kind of had\nthat rag struggle of trying to like get\nour docs to be um searchable by voice\nagents. So we have we have a voice agent\num and we wanted to answer technical\nquestions about assembly AI. So we try\nand feed off all of our docs uh into\nlike a database and you know sort of\nsimilar to the story that you were\ntelling earlier about how you know\nsetting that up can can be difficult and\nyou put a lot of effort in and you don't\nreally get that great of a result but um\nyou know it's cool to see that you know\nneedle for example makes it easy for\neven folks in marketing or product\nmanagers to quickly spin up an\nautomation and um you know uh give an\nagent access to their files. was within\none node whereas in N8N it would be a\nbit more of a more complex setup with\nmultiple nodes.\nYeah, I think we strive for simplicity,\nright? I mean it's a bit like maybe you\nhave WhatsApp came out, right? And\nWhatsApp now is kind of the text\nmessenger for everyone. But there was\nSMS before, there was like uh other\nservices before. But it's just that the\nUIX and the UI, it was so easy to send a\nmessage to another person that this kind\nof like simplification of making a\nproblem very simple and just\nunderstanding user behavior and creating\na great UIX uh I think is like sometimes\nalso very strong. Um surely the must\nalso work well. The product must also\nhave quite high quality. But I think\nunderstanding user behavior, user flow\nto then bring it like a solution which\nmakes them feel good in a way. Um\nsomething very important, right? Because\neverybody these days does not have so\nmuch time and I feel attention span is\nalso reducing these days. So some people\njust like I [ __ ] just want this to\nwork like I don't care like make it\nwork. I don't want to deep dive in a\nrabbit hole and learn about\nsemantic search. I don't care. I just\nwant this to work. So I see this\nsometimes as a technical person myself\nsometimes hard to understand because\nit's like oh it's also nice to\nunderstand deeply what's kind of going\non and the architecture behind and stuff\nbut sometimes you just don't have time\nand you just need a solution and let's\ngo you know then I also understand for\nexample I don't know so well about your\ntechnology behind assembly I voice to\ntext probably have some really cool uh\nmodels there and look cool training and\nstuff like that um but I just need a\nnote in my workflow that converts my\nvoice to text and so I mean I also would\nlove to dive deeper I find this a\ncurious topic but I also have so many\nthings to do that I don't have time to\ndive deeper into this topic um so in\nthis case I just use assembly as a note\nto do voice to text you know\nand it's nice yeah\nyeah it's like you know the there's\nthere's um there's so many things to\nexplore there's so many possibilities\nnow with vibe coding you know all of us\ncan use technologies that we previous\npreviously didn't specialize in and and\nthat's also sort of a question I have\nfor you is like um do you find more\npeople are wanting to work or liking the\nuser experience of working with a canvas\ninterface to build automations compared\nto just using something like cloud code\nto like write a script or you know where\ndo you see customers kind of finding\ntheir comfort zone because as a\ntechnical person I 99% of the time would\nrather just write a script with open\ncode or something like And\nyeah, I mean that's right. I mean uh it\ndepends on the use case. For example, uh\nmaybe you have a repetitive task that\nyou want to do repetitively then and you\nwant this to run for example every hour\nor every time a new email arrives for\nyou to set up a web hook trigger then\nwith like in your code might be a bit\nannoying to be honest. Uh so that's nice\nto work use a workflow then. Um or on\nanother hand sometimes you like if\nyou're not super technical you like also\nto see visually I see it a bit like Lego\nbricks and you kind of see the Lego\nbricks and you just build with Lego\nbricks something cool and you can see oh\nhere my Lego brick broke my note is red\nso I need to fix here maybe I chat with\nthe needle chat and needle chat is\nsaying oh yeah I see your note is broken\nI need I will fix it. Uh people like\nthis visual aspect I think because they\nfeel like a bit more like I don't know\nwhat's going on. If you just maybe use\ncode and write a script and you say okay\nPython execute my py file\nI you don't really and you have just\nvcoded that you have no idea to be\nhonest what went on in this thing. Uh so\nyou feel a bit like oh no idea how to\nfix this and then it's a bit less feels\na bit less uh how do you say trustable\nmaybe for some people like because they\ndon't know at all what goes on and maybe\nthen hard to fix also longterm and if\nyou want to repetit repetitively run a\nspecific task with a trigger note or\nsketch surely you can set up a chrome\njob also with code and stuff like this\nbut I just feel like the human beings\nare somehow visual people we like\nvisuals I think for example with Slack I\nthink does an amazing job on visuals. I\ncould also not have any visuals in Slack\nand just have a cold text messages. But\nsome are really like this\nsound or this I don't know this da da da\nor typing the other person seeing these\nthings because human beings in the end\nvisual people I think um Reddit does\nalso an amazing job in this. I think\nit's a bit like this dopamine kick you\nknow for example when you create the\ncode and you just empty py file and and\nthen it doesn't feel like so it doesn't\ngive you so much dopamine I think when\nit runs green I mean it gives you more\ndopamine when you see the small boxes\nturning green and it turns green all\nover time and then oh every turn green\noh now it works that's for even\npersonally me myself I'm a coder right\nbut sometimes I just like to use a\nworkflow because I don't know it feels\nmore it feels Somehow for some tasks it\nfeels nicer to use a workflow. I don't\nknow. Yeah.\nYeah. I mean I I Yeah. I I I can see how\nthat would be um it would kind of it's\neasier to feel a reward. It's easier to\nfeel feedback from that kind of like\nthoughtful UX and thoughtful UI. You\nknow, you can tell when when where\nexactly what is broken. Whereas in code,\nit just sort of feels like a big mess\nthat you don't even know how to\nnavigate. I I I I I definitely see that.\nI guess that's honestly a great segue to\num the next question which was can you\nshow us around Neo and like show us some\nof that that UI and that UX? I'm I'm\ncurious now after I guess hearing your\nthoughts about thoughtful design.\nYeah, sure. Uh cool. Uh so I'm here with\na needle uh and so we are in the needle\nkind of like uh space and we have\ndifferent kind of departments and so for\ndifferent departments we also try to\ncreate different kind of projects and\nfor example I could go here let's say to\nuh the marketing project and then I have\nkind of like uh different workflows and\nI could create a new workflow and let's\nsay I create a new workflow I could say\nI either use a template whichever\ntemplate I want to use or I start from\nscratch. Um,\nand I can create like this workflow\nhere. And now I'm here in this empty\nthing. And now I can go on the right\nside. And I don't know yet what I want\nto build, but I kind of know what\nproblem I have. I want a workflow that\nruns every day at uh 900 a.m.\nreads my last 24 hours uh Gmail emails\nand uh summarizes them with a agent then\nsends to # generate slack. So you may do\nlike this in vibe coding too, right? You\nwould maybe tell this to cool code or\nsomething to like this. So what I wanted\nto say is like vibe automation. So in\nthis case I describe what I want to get\nbuilt and the agent goes ahead here you\nsee and then is actually building this\nworkflow for you on the left side. So\nit's like vape coding but we call it\nvape automation and so you don't have to\nyou say yo looks good uh sounds good to\nme I have done this and then you see\nschedule trigger every day at 9 as I\nkind of like uh described it. Then we\nhave here a prompt uh a Google mail\nnote. We have here an agent node and\nit's kind of like you prompt just\neverything. You don't need to maybe I\nwould like to use a different model that\ncould be chooses for mini which is maybe\nnot the model we want to use. So yeah it\njust creates this workflow kind of for\nyou and you don't have to manually go\nahead. You can also go manually go ahead\nand now I could here for example go to\nmy assembly for example uh uh transcribe\naudio note and let's say for example I\ngot an email which we where we say we\nwill say find email I don't know that\nhas audio inside then I could pass this\nto assembly and transcribe the audio and\nmaybe then I would go to the AI agent\nand say hey you transcribe the audio and\nnow I can give my agent to it. I could\neither say for example you can search\nalso the internet and maybe I would also\nsearch uh my collection. So my\ncollection I need to decide which one I\nwant to say. Uh for example I maybe want\nto search the Epstein files. We have\nlike an Epstein file collection. Uh and\nyou could say for example yo you can\nsearch internet and search collection to\nanswer this whichever comes from this.\nNow I would need to connect to assembly\nin Google mail which I can currently\ndone but yeah basically you are done. uh\nhave created kind of a workflow for\nthis. But what I could also do if I\nwould have created this connectors, I\ncould give this all the as tools which I\nI don't have yet the assembly connector\ninside here. But I could also just give\nthis all as tools to the AI agent and AI\nagent would itself you describe\nsomething like uh first transcribe\naudio to text\nwith assembly. I would need to select\nassembly as a tool. So it's kind of like\nyou know cloudbot but with a specific\nset of tools it just has access to. You\ndecide which tools it was access to. I\nwould need to create a connector and\nthen it would have access to and then I\nwouldn't need to use the assembly node\nbut I would kind of put the assembly\ntool to the agent um and it would\nunderstand first I will always use the\nassembly tool. I will transcribe\nafterwards I will to cause the search\nweb tool or search collection tool and\nafterwards I will generate the the\nanswer for that you know so it's kind of\nlike you could either put everything\ninto one agent and let the agent call\ntools or you can just also visually see\nthe boxes here and maybe I can go to one\nworkflow which I already have set up um\nLinkedIn from Zapia find LinkedIn list\nfrom Zapia this is like a deep research\nmode. Um, track define competitors\npromptation\ntrending content generation.\nH I don't know which one I should run\nright now. I have like many different\nones. I don't know which one is\ninteresting to be honest.\nBut yeah, I think I've shown you the UI.\nYou can like invite members to your\nproject, work with them together on like\ndifferent workflows. And one thing is\ncool is this collections part. Um I\nthink the thing is that you need to\ncreate collections and so you can create\nnew collection\nassembly AI for example I hit on save\nand to this collection\nyou can uh you can then go ahead and you\ncan kind of put create connectors and\nthis is what I come to the rack part. So\nthe collections are the rack part. So\nyou can create connectors here. So you\ncan make this collection available to\nhave access to HubSpot or maybe Google\nDrive, Jira, Confluences and notion as I\nmentioned Slack. So you make this data\nrack available. But you could also say\nfor example all pages from a website and\nI could for example take your website\nassembly AI you and I could say I want\nto put 800 pages you have on your\nwebsite. I want to create an assembly AI\nconnector. Um and then I could give this\nas context as I have shown you before in\nmy workflow to my uh agent and say\nsearch collection assembly AI. I could\nsay create this connector and once I\ncreate this connector it will take some\ntime. This is 800 files and then\nafterwards here we have a chat. You see\nyou could ask this chat questions or you\nuse this chat as I have shown within the\nAI agent in a workflow. So you could\nhave for example let's say you have a\ncustomer that ask a question about\nassembly AI. you have an FAQ page\nsomewhere on your website likely then\nyou would just uh use this collection\nand you would say yo um\njust uh create like uh if if this person\nquestions uh Gmail look at the assembly\nI collection do a rack search on all\nthis data which comes here inside now we\nsee there's a lot of like data coming in\nfrom your website and generate an answer\nbased on this content that you have from\nthis collection so we have like an\nautomated customer support bot which is\nbas bas on your assembly data having\nrack already built in uh super fast. Um\nso I could go back to my workflow. I\ncreated this assembly AI workflow. I can\ngo to my agent. I say assembly AI now.\nUh I can choose this collection. You see\nI say don't search internet. Just use\nthis one and then suck. I have your kind\nof customer support with built-in rack\non your landing page data ready. You\ncould also connect like Zendes tickets\nor whatever else you have to answer\nthese customer support tickets for\nexample. and super easy to be honest.\nAll right, so that was a great demo.\nThank you so much for showing us around.\nI I um I liked how easy it was to use.\nAnd um I did also get that kick that you\nwere talking about where you can see,\nyou know, some of these bricks, they're\nyellow. Let's go in now and fix them. I\nwas uh that that was great. Thank you so\nmuch for sharing that.\nYeah, thank you for letting me share\nalso. [laughter] Yeah.\nUm, so now I guess I kind of want to\ntalk to you about being a founder and\nbuilding something, building a company.\nUm, maybe you can start by telling me\nlike how long you've been at Needle and\num, I guess the origin story where you\nstarted.\nOh, um, oh, origin story. Origin is a\nhard question. No, because like where\nwhere do you start? When you're born.\nUh, no, I mean, just kidding. Um yeah, I\nalways wanted to start a startup and I\njust thought life is a bit boring. I\nthink always I find life a bit like not\nso exciting. Uh I studied for example\ntwo bachelors at the same time because I\nthought one is not challenging enough.\nMaybe that's a bit weird and I wouldn't\nrecommend. And then I worked for some\ntime in San Francisco in a YC startup\nand I thought this is really cool and\ngreat spirit and I love love this hustle\nand drive. uh live for some time in\nSwitzerland and work there as a software\nengineer. I felt like life is a bit too\ntoo predictable. Like it's like okay,\nyou have a nice job, you have a good\nsalary and life is good, but you know, I\nfeel like life is meant to be more than\nthis. And so I wanted to create\nsomething great. I want to create\nsomething which excites me and also\nexcites other people where you can which\nis like I don't know like passion I\nguess. And then I created Needle like um\ntogether with my co-founder Uno uh\naround one year a bit more than a year\nago. And why did we start it? As I said,\nwe started as a rack API because I had\nmany friends who are software engineers,\nconsultants and so on. And they told me\nyo they may work for Accenture or Deoid\nor something like this uh or some\nsmaller consultancies\nand they said everybody nowadays wants\nto set up rag pipelines inside our\ncompany. So we are trying to use\ndifferent tools and we try to understand\nwhat can we do and I thought hm\neverybody wants to set up rack and it\nseems to be like a bit cumbersome why is\nthere not like a rack API for that like\na stand out of the box solution such as\nmaybe in the beginning you would set up\nwebsites yourself like manually or\ndatabases but then there are managed\ndatabases there are managed kind of\napplications that let you manage\nwebsites and so on and so I saw the same\nis happening with rack so rack\ninfrastructure tool\num which is the case. So, rag MPI is is\nexciting and in demand but also became a\nvery crowded market as you see like many\nplayers joined the rag API market. We\nwere very early I think back then. Um\nand then we as I said again we realized\nlike people want to automate stuff. They\ndon't want to just do the search like\nlet's say cool I index your assembly AI\nwebsite and I can do a search on this\nand can ask questions about your pricing\nor whatever. But you now need to put\nthis into like a real use case. Normally\nrack is not just used standalone like yo\nI use rack and so what I use it in a\nbusiness use case which is like customer\nsupport for example or um maybe I have a\nteam update or I want to read all my\nSlack messages and every day give like a\nsummary of what happened on Slack today\nwithin the team you know so you need to\nbring this into a business context and\nthat's where automations came to play\nvery well along I think with rack and\nthat's kind of like I think the take and\nwe raised the 2.2 2 million round. We\nare a team of like six now. Um we are\nseeing a great uh drive and uh people\nvery like excited. Um and yeah, we are\ntrying to make this like 2026 year count\nvery much and um yeah that's kind of I\nthink it but I mean story is I always\nwas excited to start a startup and I\nfeel just like bit bored if I wouldn't\ndo a startup because I feel like I think\nlife is something life is not shouldn't\nbe that predictable. I think I like that\nit's unpredictable startup times, you\nknow, every day is kind of exciting and\nit's not like you work. Yeah.\nNice. Yeah. Thank you for sharing. I\nthink um I I think that resonates with a\nlot of people like uh what you said\nabout being passionate about something\nand working on something that you really\nbelieve in. I uh that's that's that's\nreally commendable and you know I think\nuh it's it's really great that you are\nin a position where you can live that\nout every day. Um, yeah,\nyou know, you're building something and\nyou're working with customers, you're\ngiving them something that they want.\nYou know, you pay all this attention to\nthe UI and the UX. Um, that makes a real\ndifference and I'm sure it gives you\nfulfillment and um, at the end of the\nday, I guess you you're happy and and\num, that's great. That's that's that's\nwhat life is for.\nIt's the hard days, you know, not every\nday is fun. [laughter]\nThat's right. Yes.\nYeah. Yeah. Yeah. How about you share um\nsome difficult difficult times in uh in\nbeing a founder?\nI think last year in July or August. Um\nso we launched this Red Flows last year\nin October actually and we made you know\nsometimes it's like going to gym you\nneed to [ __ ] lift the heavier weight.\nUh and so we had to lift very heavy\nweights because you know the problem the\ntool we built look at the product is a\nvery is also a complex product. It's not\nlike we built a marketplace or Shopify\nstore or as you know is a you know with\nthe chat with the rack and so on is a\nvery deep tech complex product and to\nmake then on top of that like a good\nuser journey is also really hard. Uh, so\nwe really had to go very very much in\ndeep mode like hustle mode like crazy. I\nmean I'm still very much in hustle mode.\nI guess I stayed in office until 9:00.\nI'm working on the weekends and stuff\nbut uh I think back then in July I I was\nuh I think I was close to burnout to be\nhonest. I could feel it because I was\nactually I had many nightmares. I could\nlike not sleep so well. I was very\nstressed.\nOh, I think that was actually a hard\ntime. Kind of came out of that good\nfriends spending more time doing stuff\nwith friends after work and trying to\ndeconnect a bit, disconnect a bit, you\nknow, sometimes otherwise if you, let's\nsay, take your laptop into bed and you\nwork until 1:00 in the night with your\nlaptop in bed, then your sleep is going\nto be very bad and next day you're very\ntired and then again it's a safe\nenforcing effect for the next night. Um\nbut I think sometimes you also need to\ngo through these phases. Startup is not\njust like yo you ch 9 to5 and then\neverything is going to be great. You\nhave to do this extra push because\ncompetition doesn't sleep you know. Um\nother people also push. So I also\nbelieve in to work smart and not hard\nbut you also have to work smart and hard\nat the same time. I think there is not\njust like only work smart and not hard.\nSo that's kind of was a hard time. I was\na hard time. I think I was very down.\nI think the the good thing about hard\ntimes is that you come out of them\nstronger and um you know that's why\nthat's part of like the journey of being\na founder is like\nI'm not a founder. I've I've you know I\njoined assembly fairly early on. Um, and\nso I've seen people who have been here\nfor really long and um, you can tell\nthat they've been really kind of shaped\nby the struggle and um, and that's like\nthat's all part of the journey. And um,\nwhen you look back on it, I guess you\nyou have happy memories of uh, of the\ntough times because like you said, you\ncame out of it with friends and um, with\nlessons about yourself. Um, another\nthing that you mentioned was that you\nyou solve like for your own pain like\nyou had friends who were talking about\nbuilding rag internally and um you\nfigure that you might try and see how\nhow you could help. Maybe you could kind\nof tell me more about that discovery\nprocess of finding a problem to solve\nand um how you landed on rag as an API.\nYeah, I mean ra as an API was the very\nfirst tool. Uh so now we are refloa\nwhich is again listening to users right\nbut I mean it's just like I I was\nsoftware engineer myself in a company\nwhere we wanted to roll out something\nwhere we need rack API and then if your\nfriends tell you the same I'm like yo\neverybody working now on this seems to\nbe like a market uh then I followed a\nlot Reddit I think Reddit by the way\ngreat tool for market research because\npeople are honest on Reddit they don't\nlike lie they are just like uh telling\ntheir problems they are saying what they\nwere built and I saw there's so many\npeople who who had this problem and then\nuh it was just like I would say that's\nhow we validated it. Then we try to ask\nsome people hey would you use a rack API\nor we say if we have a rack API you have\nto spend less time to set the rack would\nyou use it and people were yeah I would\nuse it and then then that's kind of how\nwe validated first people trying the\nproduct using the product there was\ndemand for it and\nthen we just said let's [ __ ] go you\nknow I think sometimes you also say need\nto say let's [ __ ] go because you\ncannot always just say maybe let's stay\na bit more inside and think a bit more\nat a certain point I think you need to\nsay I go out there I do stuff I push and\nI try to make it uh because if you never\nsay let's go then you can also never\nmake it but uh yeah it's good to\nvalidate beforehand and work on\nsomething that you like but I think\ndefinitely very validated uh problem\nspace um but I think also you don't know\nso much yet in the beginning what\nexactly solution you build you know kind\nof I like this problem space it fits to\nmy skill set it fits fits to my\nco-founder skill set. It fits to my what\nI like to work on also and it fits to\nthe problem that I want to solve and the\nclients that I want to serve because in\nthe very very beginning we were working\ntogether on something very different\nalso which is a point of uh service POS\nsystem something like sum up you scan a\nQR code on a table and you pay and I\ntalked with like 200 restaurant owners\nand I hate to talk I mean I don't want\nto they are great also but I cannot so\nwell talk with restaurant owners because\nthey way of thinking is sometimes a bit\nlike old school and then as a young\ntechnical person it's somehow very hard\nto resonate with them and if these\npeople are your clients and you have to\ntalk with them every day I would have\nburned out after half a year and I would\nnever come back to this you know so you\nneed to love also your you need to love\nalso your user you need to like to build\nsomething for your user and I think in\nour case user is relatively similar to\nme or like people I like to work with\nfor or you I think you could be user I\nthink you're a nice guy also. So I would\nlike to work with you for example, but I\nwould not necessarily like to work with\nthe 60-year-old restaurant owner who is\nvery much against tech and uh a bit like\nconservative. I could not work so well\nwith him every day and see him every day\nbut still be like smiling at him say,\n\"Oh yeah, I surely help you because he's\nhe's his user, you know, a client and\nand client is king.\"\nSo\nthen very fast. That's funny. It's kind\nof like you want to work with your\nfriends and have a customer that you\nlike you genuinely like and genuinely\nwant to get along to and genuinely want\nto talk to every day.\nYeah. Exactly. Yeah. Exactly. Because\notherwise you see them every day. You\ntalk to them every day. They tell you\ntheir problems every day. You have a\nbuck and they come to you,\nyou know. So you need to like them like\nyour user. So it's like the problem you\nsolve. And I didn't know that we would\nbuild automations back then, right? But\nnow we do. So, you always need to be\nadjustable to the market. You need to be\nfast. You need to be smart and hear what\npeople say.\nUm, if you had like one piece of advice\nfor it's it's it's a great segue, but if\nyou had one piece of advice for founders\nor product managers kind of watching\nalong, what would you give?\nTo be honest, I think find a great\nco-founder. I think that's very\nimportant.\nOh, wow. Interesting. What's your\nrelationship with your co-founder like?\nOh uh it's very good I think I mean we\n[snorts] also have days where we are not\nwhere we disagree because I think as\noftenly founders have uh strong opinions\nand also some ego. I do have some strong\nopinions and they do have some ego but\nyou also need to have because you need\nto move fast and push and you need to be\nbelieve in what you think is right. It's\nalso okay to question yourself but you\nalso need to be very confident and\nsometimes if two founders are very\nconfident they can also clash because\nit's hard then to find a middle ground.\nI think uh if I first also thought maybe\nto found alone but I think it's very\ngood to have a co-founder to because\nsome days you're low the other person is\nlow but uh you are high so you kind of\npush each other you know you push each\nother up because let's say you stay in\nthe office on a weekend on a Saturday\nuntil 8 and it's dark outside it's hard\nto do alone but if you're sitting\ntogether maybe an office maybe you order\nlunch or dinner together then kind of\nenjoy it more you know you also need to\nenjoy the ride And finding a good\nco-founder is very hard. Uh I think you\nneed to put yourself into really tense\nand hard situations\nand to understand can I see this guy on\na Saturday night while I'm not in a good\nmood and we kind of have some problem in\nproduction. And if you can say yes to\nthis question then this is a good\nco-founder for you if you have\ncomplimentary skills also and they fit\nto the product. Yeah.\nNice. Thanks for that piece of advice.\nHonestly, thanks for coming on. I I\nreally enjoyed our chat. I enjoyed um\nlistening to your story. I'll be sure to\nshare like your links in the\ndescription.\nAnd yeah, everyone, that was Yan from\nNeedle. Thanks for listening.\nThank you very much, everyone. And\nthanks, Matt.",
  "transcript_chars": 35171,
  "ingested_at": "2026-05-15T16:33:20.369862+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 305,
    "like_count": 10,
    "channel_id": "UCtatfZMf-8EkIwASXM4ts0A",
    "categories": [
      "Science & Technology"
    ],
    "tags": []
  }
}