aacash.eth - Aakash Kumar

@contextconor yep, the world is just a museum of passion projects that made it out of the notes app :)

Today, we’re opening up our in-house data collection and enrichment infrastructure to all → Pre Revenue Agents: $3.1B a year spent to get contact info, titles, and firmographics → Post Revenue Agents: Agents do it for you and collapse SaaS data and enrichment tools Same top-tier providers the industry resells, at cost with no markup. Check it out. @rox_ai

As we build AI-first products at @Razorpay we wanted the interface to do more than just show answers or automate tasks. It should guide, respond and feel more intuitive and lively without feeling mechanical. That’s what led us to RazorSense, our design language for the AI era. Built around motion, context, emotion and the Razorpay glyph, it gives our products a sharper, more expressive way to communicate with customers. AI-first products need a new visual language. This is our attempt at building one. See it in action: razorpay.com/razorsense/

Everyone in the Bay is building a robot company. Almost nobody is talking about the real bottleneck. After 2 years running the Robotics Center of Silicon Valley, seeing hundreds of frontier labs and stealth startups, one pattern is undeniable: The blocker isn't model capability. It’s the sim-to-real loop. 1. Simulation ≠ Hardware Reality A VLA trained in Isaac Sim or MuJoCo hits a wall when it encounters real-world actuator limits: thermal throttling, joint backlash, and sensor drift. Even two units of the exact same robot model have completely different dynamics. 2. Teleop data is inherently off-policy Human demonstrations bootstrap your model, but your model isn't human. That distribution gap doesn't disappear with more of the same data. It only closes through real-world rollouts, failure capture, and targeted data collection. 3. "Working" is undefined until deployment Teams optimize for lab success rates, then fail on real-world edge cases: a wrinkled shirt, 4 PM glare, or a pallet 3cm off spec. Without a real-world evaluation harness, you are just guessing. 4. The solution? Treat RL² as your core loop Internally, we call this Reinforcement Learning in Real Life. Deploy ➔ capture failures ➔ evaluate against real criteria ➔ collect targeted data ➔ redeploy. Continuous learning on real hardware. 5. The Takeaway The winners in Physical AI won't just have the biggest models. They'll be the teams that can close the real-world learning loop the fastest. Closing the sim-to-real gap is a 2-year infrastructure problem. That’s why we built our stack at the Robotics Center. 6. Come hang out If you're hitting these exact walls, let’s compare notes. We host robotics builders at 90 Welsh St (SF) every Friday. Drop by.

VCs: this job is investing in outliers. Also VC: here’s our revolutionary sourcing algorithm, custom built by our in house team, RLed on decades of data, with proprietary data signals from the moment someone is born. This job is very weird where there is so much serendipity and luck involved in how you stumbled upon an investment. However, it does take a lot of conviction and if you have years of bets where you have been right, then you aren’t simply “lucky”. Returns are the only alpha.. but those take time and so we tend to resort to random metrics to prove how we’re different from the next VC. The unscientific way I have tried to play this game is by massively expanding my surface of luck to try to find these outliers generally and having a prepared mind for 3-4 areas that are 6-12 months early where I can meet every entrepreneur. Time will only tell if this works!

This video from @spencerpratt was so incredibly good

we're hiring a developer relations engineer at @dodopayments. Bangalore, in-office. our product is an API. the developer experience is the product experience. we need someone who can bridge the gap between our platform and the developers building on it. what the role looks like: → build sample apps, starter templates, and reference integrations → write quickstarts, API guides, tutorials, and SDK docs → own the developer onboarding journey and reduce time-to-first-integration → be the technical voice in our community. X, Discord, GitHub, forums → triage developer-reported issues and partner with engineering to fix them → dogfood our APIs and SDKs. catch the friction before customers do 2-5 years. hands-on technical. you can build a working integration against an API and explain it clearly. 50,000+ builders use Dodo Payments. 190+ countries. 80+ currencies. 40+ payment methods. the docs you write and tools you build are the first thing every customer touches. apply: jobs.pyjamahr.com/dodo-payments/…

@venturetwins True. Across all of g products now.

Shocked to say this, but I’m finding the AI Overviews in Google search to be incredibly useful now.

We are in the "what is ecommerce? Its just a website" phase of "RL is easy" In the immortal words of Sachin Bansal "Tu karke dikha"

Today there are unbelievably EXCEPTIONAL founders, founding EXCEPTIONAL companies that will be historic and the demand for them + their valuations will be justified BUT these EXCEPTIONS are––EXCEPTIONS 99% of co's funded today at unjustifiable prices will ERASE investor money

@8fundraising Lock

I turned #Claude into a full fundraising team Investor sourcing Warm intro mapping Pitch deck builder Cold outreach that gets replies Investor CRM from first touch to wire 475 prompts + 550 vetted investors (family offices, fund of funds, VCs). Comment "LOCK" for the full guide

I wrote this because “loop” is about to become one of those AI words that gets used for everything. That would waste the useful part. A good loop has a goal, a check, a revision step, a stop condition, and proof. It should show what changed, why it changed, how it was checked, and what a human can trust. A lot of AI work already happens this way manually. Someone asks for output, reads it, corrects it, asks again, and keeps going until the work becomes usable. The repeated correction is the signal. That is the moment a loop can take shape, and the reason I started the article there.

@ayushagarwal @dodopayments @Cloudflare 💥

a few days ago i posted about building Sentra for @dodopayments docs over a weekend. someone asked if we'd open source it. we just did. introducing context-chat. an embeddable, self-hosted AI chat widget for your documentation. one script tag. one @Cloudflare Worker. RAG-powered answers with inline citations. how it works: → pair it with contextmcp (our open-source doc indexer) or bring your own search backend → single Cloudflare Worker serves the widget and handles the chat endpoint → React widget bundled into one IIFE, mounted in Shadow DOM. doesn't break your site's styles → streams OpenAI answers with source citations → layered abuse protection: Turnstile, per-IP rate limits, global daily token/request caps via Durable Objects, AI Gateway spend limits embed it with one script tag: set window.ContextChat with your endpoint, assistant name, and starter questions. add the script. done. Apache-2.0. deploy on your own Cloudflare account. fully self-hosted. you control the data, the costs, and the model.

Land on a d2c page > Checkout powered by @gokwik makes it seamless with just an OTP > @Razorpay for PG > pay using @getscapia cc over upi > Done! 10s of slick experience delivered by friends and founders who we at @z47_vc have been privileged to back! If you are building an essential piece for the consumption economy, I’d wanna be your friend too. DM!

@tejeshwi_sharma I diagree to agree with that!

If you are eager to learn, find the smartest person who disagrees with you the most and make friends with them. It's v annoying but powerful. And even better, invest in that person. That's the hedge against your own self :)

@amnigos Think we have entered middle game now. So not really Gen1 :)

Enterprise AI is still Gen 1. The real prize is Gen 3, around 2030. Haven't started yet? Relax - you're early, not late :)

Being both of those guys at the same time is the story of my life. I am sure it’s not financially prudent 🙈 but f*** it #compulsivecollector

Physical AI data collection is 1/2 crypto grifters back with a new “plausible” con! Total 2021 redux

Excited to advance the adoption of stablecoins! Let’s update the financial system.

Investors who make the earliest Commits are the stars of pre-seed rounds. Especially those that don't ask "who else is investing?" They are often not mentioned in the final fundraising announcement but are the spark that gives the whole round momentum.