How I Built My LinkedIn Virtual Assistant - Interview with Full Stack AI Engineer

hey I'm Max the original flow gramar if you've been busy on LinkedIn perhaps you're trying to grow it or perhaps you're getting a lot of messages to your LinkedIn you've probably wanted a virtual assistant at some point and with a lot of AI capabilities evolving really quickly there's a lot of use cases where we can build some kind of AI agent to help us do some sort of virtual assistant VA tasks in our LinkedIn and various inboxes Felix a full stack AI engineer in n's Community took some time to show me a really neat use case that he built out that helps him manage his link and help him grow especially as he's on the go one really neat thing you'll see in his video is he's rigged it all up to work from Telegram and if you're a user of telegram you know why it's way better than LinkedIn chat and to be honest it's way better than WhatsApp as well so he built a use case where he can interact with his LinkedIn from telegram this is not just for messaging either each new message that he gets added into a vector store so directly from within telegram he can ask questions like what was that one thing that Max asked me about a month ago etc he can send messages to folks as well on that note let's hear it from Felix himself hey Felix how's it

going good thank you nice to reach you Max super happy to be here and uh yeah talk about my in inflow absolutely yeah thanks so much for coming on this is a interesting use case so Felix is one of our community members I actually had the privilege of meeting out in the world at a Meetup and he was telling me about some of the stuff he's working on and I was like oo we got to show this off but before we get all into that Felix would you mind introducing yourself yeah happy to introduce myself I'm Felix like Max living in Berlin originally I studied International Management nothing to do with coding I actually took the route into coding first doing uh data engineering full stack development and only like in the couple of months I got like super excited by n ATN and yeah that's how I fully got into it and got started and now happy to be also like going to the end meetups in Berlin and exchanging ideas with people I always love to hear even when folks are just a couple months in like all the stuff that they were able to do the use case the sharing the enthusiasm energy it's very infectious so Felix could you give us a bit of cont on what you're showing today and why you built it today I want to show you like my personal AI assistant and there I built a sub agent for LinkedIn which I found very exciting because one of my new in year goals is to kind of like grow on LinkedIn and build my brand a bit and I think agents are super cool because yeah they kind of can take these decisions by themselves I would say pretty Advanced but in the end if you understand it not that advanced use case would be a lot of fun where I can Tinker around with my Linkin data and have something to show so that's kind of the idea I want to show I'll show probably something around like retrieving messages how to get messages in sending a connection request and maybe also a message to Max all goes well if the demo gods are with us that sounds awesome and Hyper relevant to me because I don't know about you but my linkedin's getting pretty nuts like the everyone's super excited about AA and talk about it but I spent quite a bit of time replying to everyone I definitely want to do because you got great questions but I'm sure a little bit of automation magic could help there so Felix could you share your

screen and show us what you got let me start by maybe explaining high level how this works is as you can see everything starts with telegram I found telegram the easiest way you know it's always available even if you're on the phone in the Metro or something you can always do a quick telegram chat so we have the telegram trigger here and the telegram reply and then yeah it's pretty much connecting to an agent and then next what I wanted to do is because I already have like a calendar agent and some other stuff I wanted to kind of build like a sub agent that only is like taking care of LinkedIn stuff with a of tasks like searching in my own LinkedIn chats scraping or like accessing information from a LinkedIn profile this is kind of like the preep it's just uh loading data and I think like a cool byproduct of this is like having control of all your data like I used post hog to for example like build my own cool dashboard so I can kind of track already have 95 new followers there was one post who went like viral kind ofish which was really good so I can kind of like nicely check in here oh yeah this post was really good 67,000 Impressions to kind of get an idea of like my post data and I think once you have the data in right like there's lots of cool things like I have with this tool like searching through chats coming up with new ones running your toone that's the story of like how to get data in and then building tools on top but maybe before I continue I don't know any questions in between first reaction this is an awesome use of post hog um it's Analytics tool to analyze like channels and platform your product why not apply that to your LinkedIn that's super cool now my one question I would have is I know the LinkedIn API can get locked down pretty heavily like you basically got to be like Jeff Bezos to get access so how are you sort of fetching that data from LinkedIn I'm using a really cool company called unile and they basically allowing you to connect with your account and then they kind of like take care of all the automation so they just fully expose the rest API and I think this is like another really cool feature about love but and it end you have to do a super cool import Cur command and then it can pretty much copy from unial API the request method into here adjust the secrets right I don't want to have the secret in the code so I changed this but then it's pretty much ready to go yeah I think another kind of like cool use case which I did a lot is like in terms of like so I think for this API it's not an issue they're actually sying all the chats for you so I can use like this pagination method and hit it quite hard but what I've done before is like also actually using you know this really cool andit end expression to use math random to do like a random inter between requests so this is like a really cool hack on how to you know space out the requests or just do a for loop with a weight for kind of like things to be a bit more human this happens all the time in a community that's so simple I've never seen someone do that and that's genius for getting kind of variability in your request set yeah and I think that's kind of like the cool thing right like being able to write code and that's actually I think also like as a coding person that's why I like an at end because it has just the right amount of abstraction like when I want to write code and do things more like I would say or P ner I can do it but then again for example building pagination has always been a pain and here it's like super easy Once you understood like the pagination things it's really cool and easy super cool okay so at this point in the workflow you've basically ingested the data and in postgress you have like a onetoone representation of the stuff that you care about in LinkedIn in your own database that's right yeah this is how you have the data one tip or something I like I actually love poess I think if you're non technical maybe poess is a bit scary because it's like a database and things but what I find like especially helpful in the data pipeline is this abstract functionality and how you guys build it um all I have to do is pretty much specify in here it should map automatically and then a column to match on so if the data is already there I simply update it if it's not there I replace it so that helps me to pushing data into the database and if it's there will update if not I will insert it that's why love using Poes and PG vector and a lot of the other things that come with it in the background I'm using neon which is another really cool company neon Tac and basically what they allow you to do is have a like serverless database they're like really easy to set up and you can do database branching so you can copy like your whole database and get a copy of it so you cannot destroy anything one of my favorite tools if you're looking for cool cheap free database neon is could be cool sweet thanks for the pro tip so I think then now that we got kind of the idea the datas into the database I built this tool called search my LinkedIn chats so the way I defined it is searches analyzes LinkedIn chat messages by generating SQL CES so I just watched your prompting uh no the V3 agent YouTube video today very cool I fully agree what you said on here Jason schema has been for me like the best and that's what I've been using to just quickly use AI generate Jon schema to kind of like get the right inputs in what that subflow looks it gets it here and then what it does it's pretty crazy it generates like a srel query this can be a bit dangerous if you know someone writes like oh do a delete data drop tables right or something okay exactly but since I'm the one checking with it you could have like even right like hard filters after this and say hey if there's the word drop delete or something in there then put an error or something so I think there's always cool ways to mitigate this but obviously since it's me I take the risk this is like a basic LM chain that kind of generates the cury and I think here in the message I gave it like a good idea I created this viewer that stitches together all the chats I gave it some guidelines on how it should do filtering using like I like and wild card characters to find things that's how like yeah what were my last few messages with Max and it should summarize it for me quick question for you Felix any prompting tips maybe specifically when it's around SQL or generally things you just picked up from your experience so I think again what you said in the video I love Claude sonnet's workbench generating prompt and improving prompt it's for some reason doing fantastically well for the things that generate I think what you covered like pretty well I like to use XML text when it makes sense to give it structure here in this case I didn't but yeah I think generally just prompting around I love to kind of like use just CLA as a starting point because sometimes I also struggle like oh I need this perfect promp how do I get it perfect and you're like in paralysis you don't know how to start so I just start off with CLA and then you know Tinker around and I think that's also what I really love with in it end to be able to you know iterate click generate and you know change it generate like this so what happens next so we've generated the SQL query hopefully without drop tables and what's next basically aggregating all the results into one output and then gives this back up to the Linkin agent back up here who then gives it up back to my personal eye assistant who gives it back to F so you can see it's already like pretty cool my I agent sites the LinkedIn agent should take her over the LinkedIn agent says okay let's do the search contacts then have an LM generate the search create results come back go back up until it comes up to the chat this I don't know having this kind of multi- agentic Step where you kind of like have a high level scope and then you break it down you have an agent focus on just that I can highly recommend this patent for my own experiences and basically everyone that I'm talking with perhaps in future you won't have to break it down as much with you know the prospect of AGI and stuff but today I think this is the best practice that I've seen from people actually building stuff that's actually working is to kind of scope that down have it focused because then there's basically less degrees of freedom for it to mess up yeah exactly I think this feature you guys Ed recently right if I go into the executions and get data in here like what I've been using a lot is these locks like they really help me to understand like especially the more tools you add right it get kind of like complicated and this has been like super helpful for me and then if you just start manipulating here like the input then yeah you can really test it even in smaller steps which I like a lot because what I think is also hard with agents right once you set up this agent sub agent tool call you're like oh but I just want to debug this thing right and I don't want to have the whole thing run again and that's what I also love about NN that I can stop into the different flows reload execution data manipulate it a bit and keep on iterating which in cold is much harder and that's why yeah it became such a big end in sweet well I overand that okay I think this is the

perfect juncture to show me the money let's see it working I'm excited so I wrote can you give me one send in summary on the last five chat messages I had with Max on LinkedIn let's and see a few moments later you already responded those quick yeah okay Max rected to message scheduled onine meeting share document and discuss shooting at any end office okay let's you share it this notion link with me I couldn't find it so for preparing it so I ask you to get it again so I prepared this message I remember ma sharing a not link with me can you find it for me let's see if you can actually find it this is like actually useful you know you see some of these assistants you're like oh that's cool but like I don't ever see myself doing that this is the kind of stuff that like that could take you 5 minutes like if that message happened a month ago and the demo gods are with us luckily they show like the link you sent me for preparing I think for me where really want this to be is like remember I talked with some guy on LinkedIn about something like was that about can you please shoot this quick message over so actually what I want to do now is do that and tell it please send Max a LinkedIn message thanking him for allowing me to be on the show and when I tested it got a bit confused with the tool and said oh I cannot do that but yeah let's see if it's actually able to do it okay Max L we got let's see I don't want to there's a few other messages I want to explod but yes I did here let's have a look there you go everyone y thanks for allowing me to be on the show Max there you go hey let me just give that a little heart right there perfect and kind of last I see a lot of really cool people on link that I want to connect with and I think just sending sometimes an invite was nice but I want to put in the car to kind of like also do a bit extra so that's why I created this agent who can draft a connection request based on someone's profile and then actually sent it Lis Guzman I found like his LinkedIn link here and let's see if um able if he's able to draft a nice request for him I think kind of also what I built here like it should be able to actually pull his profile data so it's not it shouldn't be just like a generic I would like to connect but it should be really based on his profile and yeah this is what you can see here Halo saw youre of marketing at NN cool stuff then AI Big Marketing would love to connect and now if the last demo of Master works and I it all right uh perfect plan um yeah and I think it did it let's check LinkedIn real quick to like verify that it did and if I go here to my network and then manage I can actually see in the one here you can see I send it to amazing that's super cool I'll put a good word in with the big man make sure he accepts that connection request so I just my first reaction what I think is really cool and adds a little bit of I'm guessing sophistication to this as well is where it sent you the draft and then you replied in human language yes that draft looks good and this is a trend I'm seeing is where you know I think some of the first AIS we were kind of just like having buttons and stuff and adding UI and now I see folks more and more like making it feel like really a colleague that you're like messaging on slack or on telegram like an assistant would be like hey how does this work that I just did look and you're like yeah sounds good and you just send that yeah and I think yeah you're tot right like making it as a tool right and put it into an agent yeah you can like Tell it on a slightly other wording or I think also thing I want actually add an extent is should tell LM do it under 300 characters right say they do it but they never do so actually you know you could even build a tool and tell hey use this tool to check the length would just be like a code subflow like measuring the exact length of characters and then you know will iterate on itself and I think before that right like you said you had to build this complic at if then this flow Loops or something and now it's like yeah all much easier absolutely it's the slack tool as well that made me see you know a lot of my first workflows was like do the work and then like an if no check and then slack message but seeing like these AI has been more and more autonomous where it can just send that message itself or send a hey it didn't work out with relevant context etc so I definitely think it's early days in what we can build with AI agents Felix what's next

um for your agent and have you given it a name right now it's all under the umbrella of Felix personally yeah very creative so I think like long term I would love to build a dedicated agent for all of my subtasks so I already got the scheduling of appointments handled I got LinkedIn handled and yeah I think then from there right accounting maybe doing invoices or like all this tedious tus that I can imagine I think crossposting is something also like a very interesting thing already worked on this morning so I think soon another will come Felix first off thanks so much for your time on this one and it's really cool to see again I've seen some of these personal assistants but this one really seems like it's a good place in terms of like being iterated a bit the way you're thinking about it the way you're laying it out to be able to build on it in future I think a lot of personal projects we hack together we don't have as much thinking on that and then we get a lot more debt but this seems like it's really set up to grow and being honest I think there's going to be some like some CEOs and some sales folks and stuff they're going to see you LinkedIn want to be like gimme gimme because I think that could be like a product in and of itself so do you plan to publish or share any of this stuff publicly work for sure I can share it I'm happy to share what I chose in it and it's like it's open source right so that's what I love about it it's all about like sharing and giving back so you know I would be happy to I'm still in the learning process of how do I do the proper template with notes and everything but yeah happy to open source parts of this you know I think like the LinkedIn data pipeline or like even sending the simple ones for it like connection request I'm happy to knew that and I think also I don't know max if you saw this but and it's like a rising star in the JavaScript ecosystem you guys got 17K stars in 2024 so you're officially on top five now in JavaScript ecos system so I think that underlines like such a cool and Rising product because I think you found like a perfect abstraction of being code being attractive to coders being also attractive to no code people I'm super excited for it and happy to share this template with you guys firstly thanks so much Felix for the kind words and like your just adoration for like what we're doing and stuff and your support I was ent's founding designing I've been working on this for 5 years right now I think what keeps me around is that kind of authenticity whole community so when you do post that template send me a quick DM and I'll do my Shilling thing and I'll let everyone know about that thanks so much Felix and I'd love to have you back on the show be it you evolving this use case or building something else so keep that in the back of your mind as you're flamming out there because I think this was really awesome and I'm sure everyone will be interested to see what else you bu sounds cool I already have one project in mind I'm happy to it has something to do with in community notes I'm already working on it so maybe that might be interesting maybe something else yeah I I'll keep you posted on the progress Felix this was awesome if people want to follow along your journey where can they do that I'm actually on LinkedIn as you one of my goals is to goow by LinkedIn so if you would love to follow me uh trying to share more and more like good knowledge there and I'm also building up my small YouTube channel only two videos there but hopefully coming Mar sweet so gang follow him on YouTube and linked if I can do a link in the video it will be if not it's going to be somewhere above below you know the drill share the love help the flow gramar out Felix this has been awesome thanks so much have a beautiful rest of your week and chat soon bye thank you Max have a good

one hey cheers Felix appreciate that one I'm always blown away at how much our community can do on their own right this use case as you saw Felix builted out himself to help his own personal goal of growing his LinkedIn he didn't have a team to build this and you know I'm sure Felix has bills and red to pay and stuff like that so he can't spend a week on this either that's I think the really exciting thing about 2025 right now and all this technology is how much you can do on your own but how much you could in an amount of time that you can justify on your own taking that one step further imagine what a team of two or three could do for a week right in a lot of organizations you can find a team of two or three for a week if you can solve a big problem like that so yeah early days in AI still right like this is stuff is only getting faster if you're a fan of the show and you've been following along you'll notice that we're just doing the interview with Felix this week this is something we're testing out single segment distribution so this video with Felix will go out but the next video that's going to go out is going to be super special because I just got back from Dubai where I was building with un Basu and we're working also on a LinkedIn use case basically an app to manage our inboxes um that should be good for SE suite for influencers for anyone that's being harassed by their followers in their inbox thank you by the way guys those messages are awesome I just want to be able to get back to all of them so yeah we're going to do this for a few weeks we're going to try out this new format I've also onboarded a video editor now ol thanks for that first edit really nice one he actually edited the video you just saw the good thing about that is we should be able to ramp up so more interviews coming out more often more feature walkthroughs in N at end and then more special projects like my trip out to Dubai where with only in 2 and 1 half days we built out a really cool solution so make sure to be following the studio playlist all these videos are going to be released on there as the studio update was and again I know change is weird change is scary but bear with us guys because in classic Scrappy startups we're innovating we're trying you things we'll reflect we'll retro and we'll see what we do moving forward so again do let me know in the comments how you like this format and if you hate it stick around for a couple more because at the end of the day it's still Max it's still flamming and being honest it makes it a bit easier for me to not have to have a weekly episode because my mental health is doing my head in uh with the weekly deliverable I'm Max thanks for watching and happy flog gramming