Pradeep

i have always wanted to build a startup since college. i bailed on IBM after graduating from IIT, joined finbox instead (grew it with the team to series B) realised i wanted to learn more about AI, so joined superAGI. took a stab at building a fitness startup by moving to dharamshala and failed. finally joined @join_ef. started building with @utsavmazumder05 and @its_shubho. while exploring various problem statements, we finally landed on something really interesting. we realised together that brands need production bandwidth, and not ideas. they want to test 50 ad concepts but can only produce 5. that's the bottleneck. so we built novelty. you bring scripts, we deliver AI video ads in 3 days. we're onboarding partners now who want to scale their video production, if you are interested lets talk?

something remote work taught me async is better than sync we work across time zones because @its_shubho @utsavmazumder05 are in SF no real-time collaboration because everything is documented. result is better decisions less meetings = more shipping

the biggest mistake i made was overengineering the first version i built microservices, kubernetes which was a complex deployment i should have built a monolith, deployed on a single server and scaled later. complexity before scale kills startups

the best startups have boring tech stacks they have proven tools and standard patterns save your innovation for your product, not your infrastructure

the best code is the code you do not write every line of code is a liability simplicity wins

engineers love solving hard problems founders need to solve the right problems there is a difference.

the best infrastructure is invisible if your team thinks about it, it is broken if your team forgets about it, it works.

We went from 10 videos per week to 20 videos per week. - No new hires - Same infrastructure Just better pipeline design Engineering leverage is a real thing.

We automated 80% of our video workflow, which we use to generate 100 ads per week. We used to handle: Generation → Rendering → QC → Delivery Now humans handle: Creative direction → Final approval Everything else is automated. Result: 1 person produces 20 videos per week. Reply "100" and I will send you the exact playbook we use with prompts.

We automated 80% of our video workflow which we use to generate 100 ads per week. We used to handle: Generation → Rendering → QC → Delivery Now humans handle: Creative direction → Final approval Everything else is automated. Result: 1 person produces 20 videos per week. Reply "100" and I will send you the exact playbook used with prompts.

Most startups I saw hire too early. Before they even have product-market fit. Then they spend time managing people instead of building a good product. Stay small but focused and hire when you cannot keep up with demand.

Most engineers want to build everything from scratch but I use libraries. For auth, I use Clerk For payments, I use Stripe For the database, I use Postgres Building from scratch can take months, but using tools can take just days. Save your engineering time for your core product.

The best engineers are lazy, so they automate everything. Why do something manually when a script can do it in 10 seconds?

The hardest part of startup engineering is saying no. No to unecessary features or to unnecessary tech or to complexity. Every yes slows you down, but focus wins.

Most engineers optimize for elegance. But I realised that it's better to optimize for speed. Clean code that ships tomorrow beats perfect code that ships next month.

Your north star metric is probably wrong. EF made us pick videos per week per person. Went from 5 → 20 → targeting 50 Focus changes everything.

Most AI infrastructure is just 10 tools duct-taped and packaged as a product.

I rejected IBM after IIT. I joined a fintech startup at half pay. Safe path optimizes for safety, not learning speed.

@Google needs to fix this before rolling out gemini everywhere. (2) We use Gemini across scripting, product research, storyboarding, and video generation, so when the rate limits started hitting hard, it wasn't a minor inconvenience, it was affecting our client delivery. Solution: - Talking to google right now - Using fal ai as third-party backup untill google resolves it permanently. The second problem is Veo blocking celebrity faces and children by default at the content restriction level, which cuts out a meaningful category of shots for consumer brand work. We're using Kling for those specific cases while we talk to Google about whitelisting Novelty. Google's fixing this, but the gap between "what the demo shows" and "what the api allows" is still the biggest bottleneck in ai production. I'm curious whether others have faced this issues, let me know!.