
Skill meets opportunity ๐ Compete in global challenges, win equity & funding. Join #Startupathon by @PersistVentures
LA United States
Joined May 2025
- Tweets99
- Following8
- Followers24
- Likes54
Everyone wants to be a founder.
Almost no one wants to build.
Startupathon = Find a challenge โ Build โ Get feedback โ Earn equity.
Comment HOW if you want the playbook.
#Startupathon #BuildInPublic #FounderLife #ExecutionOverIdeas
30
๐ป๐๐๐๐ ๐ด๐ฝ๐ท๐ ๐๐๐๐ ๐๐๐๐๐ ๐
๐๐๐๐๐ ๐ ๐๐๐๐๐๐๐
๐๐๐๐๐๐๐๐๐ ๐๐๐๐๐๐ ๐๐๐๐ ๐๐ ๐๐๐๐๐๐๐๐ ๐๐ ๐๐๐๐๐๐๐๐๐, ๐๐๐๐ ๐๐๐๐, ๐๐๐
๐๐๐๐ ๐๐๐๐๐๐ ๐๐๐๐๐๐๐๐๐๐๐.
๐๐ถ๐ฅ๐จ๐ฆ ๐ง๐ฆ๐ฆ๐ฅ๐ฃ๐ข๐ค๐ฌ ๐ฉ๐ช๐จ๐ฉ๐ญ๐ช๐จ๐ฉ๐ต๐ฆ๐ฅ ๐ด๐ต๐ณ๐ฐ๐ฏ๐จ ๐ฎ๐ฆ๐ค๐ฉ๐ข๐ฏ๐ช๐ค๐ด ๐ข๐ฏ๐ฅ ๐ค๐ญ๐ฆ๐ข๐ณ ๐ถ๐ด๐ฆ๐ณ ๐ซ๐ฐ๐ถ๐ณ๐ฏ๐ฆ๐บ๐ด ๐ข ๐ฑ๐ณ๐ฐ๐ฎ๐ช๐ด๐ช๐ฏ๐จ ๐ฃ๐ข๐ด๐ฆ ๐ง๐ฐ๐ณ ๐ง๐ถ๐ณ๐ต๐ฉ๐ฆ๐ณ ๐ช๐ต๐ฆ๐ณ๐ข๐ต๐ช๐ฐ๐ฏ.
๐๐ฉ๐ช๐ด ๐ช๐ด ๐ฆ๐น๐ฆ๐ค๐ถ๐ต๐ช๐ฐ๐ฏ ๐ช๐ฏ ๐ฎ๐ฐ๐ต๐ช๐ฐ๐ฏ.
#StopScrolling #ViralLoading #WatchTillEnd #MainCharacterEnergy
#BuiltDifferent #ThisChangesEverything #TrendingNow #FYP
#GameChanger #InternetBreaking
2
33
STARTUP/ATHON retweeted
โก Tech Tuesday: Your weekly dose of AI power.
Swipe โ Discover โ Upgrade. ๐ฅ๐ค
๐๐ญ๐๐ซ๐ญ๐ฎ๐ฉ๐๐ญ๐ก๐จ๐ง: ๐๐จ๐๐ฅ๐จ๐ฑ ๐๐ข๐๐ ๐๐จ๐๐ข๐ง๐ ๐ข๐ฌ ๐๐๐๐!
๐๐ฐ๐ญ๐ฅ๐ฆ๐ฏ ๐๐ฑ๐ฑ๐ฐ๐ณ๐ต๐ถ๐ฏ๐ช๐ต๐บ ๐ง๐ฐ๐ณ ๐๐ ๐๐ฏ๐จ๐ช๐ฏ๐ฆ๐ฆ๐ณ๐ด! ๐๐ฐ๐ฅ๐ฆ๐ณ๐ด, ๐ค๐ณ๐ฆ๐ข๐ต๐ฐ๐ณ๐ด, ๐ข๐ฏ๐ฅ ๐ท๐ช๐ฃ๐ฆ-๐ฃ๐ถ๐ช๐ญ๐ฅ๐ฆ๐ณ๐ด ๐บ๐ฐ๐ถ๐ณ ๐ค๐ฉ๐ข๐ญ๐ญ๐ฆ๐ฏ๐จ๐ฆ ๐ด๐ต๐ข๐ณ๐ต๐ด ๐ต๐ฐ๐ฅ๐ข๐บ.
๐๐ฉ๐ช๐ด ๐ช๐ด ๐บ๐ฐ๐ถ๐ณ ๐ค๐ฉ๐ข๐ฏ๐ค๐ฆ ๐ต๐ฐ ๐ต๐ถ๐ณ๐ฏ ๐บ๐ฐ๐ถ๐ณ ๐๐ฐ๐ฃ๐ญ๐ฐ๐น ๐ด๐ฌ๐ช๐ญ๐ญ๐ด ๐ช๐ฏ๐ต๐ฐ ๐ด๐ฐ๐ฎ๐ฆ๐ต๐ฉ๐ช๐ฏ๐จ ๐ณ๐ฆ๐ข๐ญ, ๐ฆ๐น๐ค๐ช๐ต๐ช๐ฏ๐จ, ๐ข๐ฏ๐ฅ ๐ค๐ฐ๐ฎ๐ฑ๐ฆ๐ต๐ช๐ต๐ช๐ฐ๐ฏ-๐ณ๐ฆ๐ข๐ฅ๐บ.
๐ก ๐ฉ๐๐๐๐
๐๐๐๐ ๐๐๐๐ ๐๐๐๐
๐งฉ ๐บ๐๐๐ ๐๐๐๐๐ ๐๐๐๐๐ + ๐๐๐๐๐๐ ๐๐๐๐๐๐๐๐๐๐
๐ฅ ๐บ๐๐๐๐
๐๐๐ ๐๐๐
๐๐๐๐ ๐๐ ๐๐๐ ๐๐๐๐ ๐๐๐๐๐
APPLY: persistventures.com/ongoing-โฆ
๐๐๐ ๐๐ง๐๐ฃ๐ ๐๐จ ๐ค๐ฅ๐๐ฃ.
๐๐๐ฉโ๐จ ๐จ๐๐ ๐ฌ๐๐ค ๐๐๐ฃ ๐ซ๐๐๐ ๐๐ฃ๐ ๐๐ค๐๐ ๐๐ฉ ๐ฉ๐๐ ๐จ๐๐ข๐ ๐ฉ๐๐ข๐.
#Startupathon #RobloxVibeCoding #RobloxDev #VibeCreators #PersistVentures #BuildTheFuture
2
26
Every subsidiary has its rhythm. Todayโs profit pulse proves ours is beating stronger than ever. Growth, grit, and great people behind it. #PersistVentures #ProfitPulse
2
93
๐ญ Every founderโs battle is different.... co-founders, first users, sleepless nights, or that one โyes.โ
๐ฅ At Startupathon, we donโt talk about it... we build through it.
๐ Drop in the comments whatโs your biggest founder struggle right now?
#Startupathon #FounderLife #BuildMode #PersistVentures #StartupJourney #MVPtoMarket
1
29
STARTUP/ATHON retweeted
AI isnโt replacing humans.... itโs replacing habits. Those who learn how to think with AI will always lead those who donโt. โก๏ธ #AI #FutureOfWork #startupathon
The next billion-dollar founders arenโt using AI tools
theyโre building on AI frontiers:
๐ง Generative ยท ๐ฌ Multimodal ยท ๐ค Agentic ยท ๐ Edge
๐ Synthetic ยท ๐ฌ Emotion ยท ๐ก๏ธ Cyber ยท ๐จ Design
๐งฉ Neuro-Symbolic ยท โ๏ธ Quantum
#Startupathon #AIFrontier #BuildBold
๐ก Fail Smart. Build Bright.
Every crack shows where to grow stronger.
At #Startupathon, failure isnโt the end... itโs the blueprint.
Founders here donโt hide from mistakes... they engineer from them.
Thatโs how early setbacks evolve into success stories.
๐ฅ Because true innovation begins where fear ends.
#FailSmart #FounderMindset #BuildBright #StartupCulture #IdeationPhase
1
28
๐ 6 Hidden Free AI Tools Every Startup Should Know
Forget ChatGPT & Notion... these underrated AI gems help founders:
โ๏ธ Automate workflows
๐ง Summarize meetings
๐ฌ Capture leads
๐ค Manage clients
All 100% free.
๐ Discover your new startup superpowers.
#Startupathon #HiddenAIGems
2
61
They didnโt wait for a degree to start... they built one through experience. ๐โก๏ธ๐
~Some of the most successful founders didnโt finish college... not because they quit learning, but because they started learning differently.
~They chose execution over expectation.
~Growth over grades.
~Mindset over mandate.
In this post, weโve highlighted a few stories of college dropouts who built category-defining companies... not by luck, but by thinking long-term and acting bold.
๐ Swipe through to see what happens when mindset beats conventional paths.
#MindsetMatters #Entrepreneurship #StartupStories #FoundersJourney
3
35
Different founders. Different journeys. One mindset. ๐
The real ones choose:
๐ Vision > Comfort
๐ Learning > Ego
๐ฅ Consistency > Hype
These arenโt just stories... theyโre reminders of what it takes to build something that lasts.
#StartupMindset #Leadership #Growth
3
30
You donโt need permission to start.
Most founders wait for โthe right time,โ โthe right team,โ or โthe right idea.โ
Truth? The right time is when you decide to move.
๐ Start small. Start messy. Start now.
You can fix execution. You canโt fix inaction.
๐ What was your โIโm doing this no matter whatโ moment?
#startup #startupathon #NATASHA #founders #Community
1
37
๐ต From a hackathon spark to a full-blown startup!
SongGPT the AI that transforms your videos, images & text into original songs was born at #Startupathon.
Where creativity meets code, and code creates music.
๐ Turning imagination into sound.
#AI #MusicTech #Innovation #Startupathon @songgpt
4
61
RIP Fine-Tuning: How Stanfordโs ACE Framework Teaches AI to Learn Without Retraining #stanford #startupathon #explorer
Stanfordโs ACE shows you can make a model learn without touching its weights. The trick is to treat the prompt as living memory and update that after every run.
You start with a normal system prompt and a base model. The model tackles a task and produces a full trace of what it tried and why it thinks it worked or failed. A second pass acts like a coach: it rereads the trace, points out mistakes, names the fix, and turns that into a small, concrete rule or example. A third pass is the librarian. It decides whatโs worth keeping, merges it with what you already know, and writes a tiny patch to the context instead of rewriting the whole thing. Next task, you load the base prompt plus those patches; so the model begins the day smarter than it started.
Those patches arenโt just fluffy tips. They can be action macros for agents (โif 429, backoff and retry with new headersโ), domain heuristics for reasoning (โwhen parsing 10-Q tables, normalise quarter labels before comparisonโ), or bite sized exemplars that show the exact inputโoutput you want. Because ACE appends focused deltas rather than replacing the whole prompt, it avoids the usual โprompt rotโ where rewriting shaves off useful detail and accuracy slides.
Itโs self supervised from the task itself. No labeled dataset, no gradient step. The โReflectorโ grades with simple success signals (did the page parse, did the API return the right shape, did the unit test pass) and with cheap critique prompts. The โCuratorโ keeps the context under a hard token budget by deduping similar rules, compressing them, and retiring stale ones. If a new rule conflicts with an old one, it logs the conflict and prefers the one that actually led to success. Think of it as online RL lite with human readable memory.
Why this beats one off prompt engineering is compounding. A single clever prompt can lift a model for one task, once. ACE collects micro lessons across thousands of runs and turns them into a reusable playbook. It also travels: the same memory can be mounted on a different base model, swapped out if it misbehaves, or rolled back when a patch hurts performance. That reversibility is a big deal; unlike fine tuned weights, you can audit exactly which rule changed behaviour and why.
The gains in the paper are reflected in numbers and in latency/cost. Because youโre not doing weight updates or long chains of external tools, you get faster rollouts and lower spend while still improving success rates on agent benchmarks and finance reasoning. In practice, that means cheaper, quicker iterations with visible learning rather than a week long fine tune that might or might not generalise.
Give an agent a job like โbuild a Slack style chat appโ or โscrape and compare vendor quotes.โ First day it fumbles on pagination and auth. ACE captures that failure, distills โalways paginate via nextLink; refresh token when 401,โ and pins those rules in context. Day two, it never makes the same mistakes. A week later the playbook includes retries, schema validators, and safe tool use patterns; all learned from doing the work.
There are guardrails. If you let the context bloat, retrieval slows and quality dips, so you need pruning and clustering. If you accept every critique, you risk locking in superstitions, so you gate updates behind measurable wins. And you still want tests; ACE learns from outcomes, so good outcome checks are the backbone of good learning.
The headline isnโt โprompts are magical.โ Itโs that structured, incremental memory turns a static LLM into a self tuning system. Weโll still fine tune for raw capabilities and efficiency, but a lot of day to day improvement will come from smarter context: patches not gradients, playbooks not folklore, and models that remember.
3
69
Thatโs next-level learning! ๐ฅ
AI turning platforms like Coursera into interactive mentors is exactly what the future of education looks like.
Love seeing this kind of innovation! ๐
2
44
๐ Sora 2 is changing the game.
From text to ultra-real videos with synced audio, itโs not just AI video itโs a creative revolution. ๐ฌโจ
#AI #Sora2 #Innovation #OpenAIDevDay #startupathon #NATASHA
2
146
If AI were human, theyโd never sleep just keep โupdating.โ ๐ค๐ค
Theyโd wake up smarter every morning. ๐ก
And still say, โSorry, Iโm in beta.โ ๐
#aivideo #funny #explore #StartUp #startupathon
4
47


































