Transcriptly

Built to Share: 2M Users in 18 Months | Julius AI, Rahul Sonwalkar

EO09:59Added Aug 31, 2026

This episode features Rahul Sonwalkar, founder and CEO of Julius AI – the leading AI data analysis platform.Since launching in 2023, Julius AI has helped use...

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. Intro

  2. 00:00I quit my job and then it took a year
  3. 00:01and a half for me to launch Julius after
  4. 00:04that. When I when I went all in, I was
  5. 00:06nervous to be honest. I was scared, but
  6. 00:08it's being a little scared is good. I
  7. 00:10think fear is good. You It keeps you
  8. 00:12sharp. It keeps you on your toes. You're
  9. 00:14able to see through things. When you're
  10. 00:17in a comfortable place, it it breeds
  11. 00:19complacency and you're often, you know,
  12. 00:22fooling yourself into believing that
  13. 00:23your ideas are going to work one day.
  14. 00:25But then when all the bridges are burned
  15. 00:27and you really have to make a thing
  16. 00:28work, you start thinking, you start
  17. 00:30taking risks, taking shots at goal. But
  18. 00:33having no other boats, you know, no life
  19. 00:36boats, it it was a little scary, but I
  20. 00:38think it was good.
  21. 00:44I'm Rahul, founder and CEO of Julius AI.

  22. About Julius AI

  23. 00:47Julius is an AI data analyst. If you
  24. 00:49have data on your hands, Julius will
  25. 00:51help you get insights from your data
  26. 00:52within seconds and help you make charts
  27. 00:54and data visualizations. Since launching
  28. 00:56in 2023, our users have used Julius to
  29. 00:59make over 10 million data
  30. 01:01visualizations. Every day, Julius writes
  31. 01:04over 4 million lines of analysis code. 4
  32. 01:07million lines of code is more than an
  33. 01:09army of data scientists can write every
  34. 01:11day. It took about a year and a half to
  35. 01:13get to 2 million users.

  36. Rule No.1 Focus on only one task

  37. 01:22For AI startups, I'll say focus. Don't
  38. 01:26build a general purpose tool. Startups
  39. 01:29win because of focus. That's the only
  40. 01:32advantage you have as a startup. Now,
  41. 01:33Chad GBD is like all-in-one tool. Making
  42. 01:36videos, making images, doing internet
  43. 01:38search, writing essays, and it's just
  44. 01:40allin-one tool. And it's not focused on
  45. 01:43data analysis. So, the experience is
  46. 01:44much worse. A lot of our users they
  47. 01:47start with using Chad GPT and they
  48. 01:50realize that to analyze any kind of real
  49. 01:54data to get any kind of meaningful
  50. 01:56insights Chad GPD falls short. The
  51. 01:59quality of insights you get aren't that
  52. 02:01deep. It cannot handle any kind of real
  53. 02:03data. The charts don't look good. You
  54. 02:05can't collaborate with your colleagues
  55. 02:06on the analysis. So all these problems
  56. 02:09people run into after using it for a day
  57. 02:11for analysis and they all go to Google
  58. 02:14and they look up AI data analysis and
  59. 02:17the number one search result on Google
  60. 02:19for that is Julius.
  61. 02:22It's kind of like think about hiring
  62. 02:23humans. Would you want to hire one human
  63. 02:26that can do everything? You you can and
  64. 02:28that will help you get decent results.
  65. 02:31But then there's comes a point where you
  66. 02:33want someone with deep expertise,
  67. 02:35someone who's really really good at one
  68. 02:37function like a marketer, finance prof,
  69. 02:40finance person, an engineer. I think
  70. 02:42focus agents for a task have much more
  71. 02:46competitive advantage than like the
  72. 02:47general purpose agent.

  73. Rule No.2 Solve daily problems that matter

  74. 02:54You know, there's this like conventional
  75. 02:56meme like Google just killed your
  76. 02:57startup, OpenAI just killed your
  77. 02:59startup, XYZ just killed this all
  78. 03:01startup. I think all that is very
  79. 03:02overblown. As long as your users don't
  80. 03:05care about that stuff, it shouldn't
  81. 03:06matter. If you're building something
  82. 03:07that solves a problem for people, if
  83. 03:09you're doing that good job at that
  84. 03:10better than everyone else, that all
  85. 03:12that's all that matters. So, this is
  86. 03:14back in college when I got into the
  87. 03:16hackathon scene. I would go to these
  88. 03:17hackathons. The big problem I noticed is
  89. 03:20you have 48 hours to build and launch
  90. 03:22idea. A lot of the time people would
  91. 03:24spend on setting up a backend service,
  92. 03:27setting up their database and I thought
  93. 03:30that was a waste of time. So water
  94. 03:32review was this, you know, managed
  95. 03:34service. You would get out of box like a
  96. 03:36backend server or database that just
  97. 03:38kind of worked. It got a lot of users
  98. 03:40and I would go to these hackathons, give
  99. 03:42it to people and they would try it. It
  100. 03:44would save them time. But problem was
  101. 03:46none of the people that built their
  102. 03:48hackathon projects, their weekend
  103. 03:50projects on waterview continued to work
  104. 03:51on those projects after that weekend.
  105. 03:53You know, it was like a weekend thing.
  106. 03:55The big lesson there was you can solve
  107. 03:58the right problem. You can solve a
  108. 04:00painoint for people, but if they don't
  109. 04:02have that pain point daily or weekly,
  110. 04:05they're not going to retain. They're not
  111. 04:06going to come back to the product. You
  112. 04:08you have to solve a problem people have.
  113. 04:09If they have it only once a year or once
  114. 04:12every 6 months or once a month, they're
  115. 04:14not going to retain to your product.
  116. 04:16That's the reason why water we failed.
  117. 04:18And that was a very valuable lesson.

  118. Rule No.3 One ‘NO’ kills an idea

  119. 04:25I worked as an engineer at Uber and
  120. 04:27Facebook. When you're at a big company
  121. 04:29and you have an idea, one no can kill
  122. 04:31that idea. Get a yes from your manager.
  123. 04:33You have to get a yes from your
  124. 04:34manager's manager, design manager,
  125. 04:36product manager. You have to get all
  126. 04:38these yeses and even one no can kill
  127. 04:41your idea. So innovation doesn't really
  128. 04:44happen in big companies. And then at a
  129. 04:46startup it's complete opposite. All you
  130. 04:48need is one yes. Looking for customers
  131. 04:50you know you can get 50 nos but one yes
  132. 04:53your first customer that's all that
  133. 04:54matters. Instead of one no killing your
  134. 04:56idea. One yes can really make your
  135. 04:58company work. So, one of the problems I
  136. 05:00was trying to solve at Uber is people
  137. 05:03book Uber rides usually for things that
  138. 05:07they can't drive to, like uh like an
  139. 05:10airport or from an airport or when
  140. 05:12they're out going out at night to bars
  141. 05:15or restaurants. That sporadic usage is
  142. 05:18not good for Uber. One of the things I
  143. 05:20wanted to solve for is how could we get
  144. 05:22people to use Uber on a daily and weekly
  145. 05:24basis? So, I wanted to launch this
  146. 05:26commuter product at Uber. How can we
  147. 05:28help people use you know Uber's
  148. 05:30offerings Uber transit Uber X Uber um
  149. 05:35Uber mobility like bikes as scooters all
  150. 05:37that as a part of a commuter package
  151. 05:39that companies could offer their
  152. 05:40employees. So I wanted to pitch the
  153. 05:41idea. I was pushing it really really
  154. 05:43hard and you know I got buy in from a
  155. 05:46lot of cross functional management. I
  156. 05:48couldn't get behind from our um
  157. 05:50engineering leader and I couldn't get in
  158. 05:52from the engineering leader and that
  159. 05:54crush idea but at the same time I was
  160. 05:55building apps and different ideas on the
  161. 05:58side on the weekends and that motivated
  162. 06:00me to just like quit my job and spend a
  163. 06:03year exploring my ideas full-time.

  164. Rule No.4 Failing Fast

  165. 06:11Every failure is is is good. Failing is
  166. 06:15good because you know what's what's not
  167. 06:18working. I I I talk to founders and they
  168. 06:21they're too scared to launch, too scared
  169. 06:23to tell people what their idea is.
  170. 06:25They're too scared to put the product
  171. 06:26out there. I'm the opposite. You know,
  172. 06:28at Julius, we launch things when they're
  173. 06:30barely working because you want to get
  174. 06:32early feedback from our users, our
  175. 06:34customers. We want them to try it out.
  176. 06:35We want to learn from our users and
  177. 06:37customers on what they actually want and
  178. 06:39how the things should work. So, I'll
  179. 06:40give you one of the examples. We were
  180. 06:42building in you know data and AI and
  181. 06:44insights and we thought you know one of
  182. 06:47the one of the group of people that want
  183. 06:49insights from data but don't have the
  184. 06:51expertise to go do data science on their
  185. 06:53own is is sports fans. We built this
  186. 06:55thing called NBA GPD was called hoops
  187. 06:58GPD could help you query NBA data
  188. 07:00playby-play data but just asking simple
  189. 07:03questions. We had to build the UI, the
  190. 07:05interface, figure out how this text to
  191. 07:08querying engine is going to work and
  192. 07:11ship the whole thing in like two weeks
  193. 07:13because if you were too late, the NBA
  194. 07:15season would be over. It was no point to
  195. 07:18wait a year to try the idea again. And
  196. 07:21turns out like sports fans aren't that
  197. 07:24savvy about data. Turns out it's only
  198. 07:26the people that want to do betting. Yep.
  199. 07:28Really aren't the users you want to
  200. 07:29serve. So we learn very quickly know
  201. 07:32whether it's something people want or
  202. 07:34don't want. We don't want to build
  203. 07:35something that no one will ever want to
  204. 07:37use. I think failing is important.
  205. 07:39Failing fast is really really important.
  206. 07:41Paul Graham talks about this in his
  207. 07:42essays all the time. It's like failing
  208. 07:44fast is super important.

  209. Rule No.5 Build products people share

  210. 07:50This is months after launching Julius.
  211. 07:53We had about 10 15,000 users and most of
  212. 07:56our users at that point were actually
  213. 07:58coming from Chad GPT. Chad GP is this
  214. 08:00plug-in store where users would discover
  215. 08:03Julius and they would come to Julius,
  216. 08:06stay with Julius and use Julius over and
  217. 08:08over and it's going great and it's going
  218. 08:10great. We're growing overnight. Open
  219. 08:13announces that the plug-in store is
  220. 08:14going to get shut down and our biggest
  221. 08:16source of users disappeared overnight.
  222. 08:19So we had to scramble and figure out how
  223. 08:23we're going to get more users and where
  224. 08:24we're going to get users from. So that
  225. 08:27was an existential moment and that is
  226. 08:30what really got us to move fast and
  227. 08:32figure out okay how do we get our users
  228. 08:34to become champions of Julius. Then we
  229. 08:37realized that when people analyze data
  230. 08:40they want to tell their colleagues about
  231. 08:42it. When you get an insight you don't
  232. 08:44keep it to yourself. You want to tell
  233. 08:45your colleagues you want to tell your
  234. 08:46team. And so we build sharing into the
  235. 08:49product. If you solve a problem that
  236. 08:51people have, they will come to you.
  237. 08:52They'll come to you and they will use
  238. 08:54your product and they'll tell their
  239. 08:56friends about it. So word of mouth is a
  240. 08:59free way to grow your product.
  241. 09:03If I were to start from scratch today,
  242. 09:05what would I do differently?
  243. 09:09Nothing. I think you know all the things
  244. 09:12I would consider as like missteps things
  245. 09:16um I could have avoided I think those
  246. 09:18are valuable lessons you know all the
  247. 09:21features we tried and didn't work out
  248. 09:23all that is really valuable lesson
  249. 09:25that's helpful data you can if you have
  250. 09:28only data from successes then that's
  251. 09:30very skewed you know what doesn't work
  252. 09:33so it's really important to try a lot of
  253. 09:34things things that don't work also
  254. 09:37important because now you know what
  255. 09:38doesn't work honestly like I would not
  256. 09:39do anything differently.