Founder & CEO of Cyborg Network — building distributed AI inference infrastructure for real-time machine-to-machine applications. I work at the intersection of edge computing, low-level systems, GPU optimization, and blockchain-backed integrity.
Currently experimenting with ways to make AI deployment as seamless as web deployment. Also building Oxlo, a global platform that lets developers launch and test AI models on GPU instances (free tier for small models, like Vercel but for AI).
Background in Rust, Substrate, Linux systems, and computer vision pipelines. Focused on building practical infrastructure that scales — from decentralized edge nodes to secure on-prem inference for smart cities.
Always interested in discussing: distributed compute, model serving, real-time AI, hardware acceleration, and deep-tech company building.
What’s interesting here is Nvidia leaning into end-to-end “reasoning” models for driving instead of ever more brittle rule stacks. If this actually fuses perception, prediction, and planning into a single learned system, it could reduce a lot of hand-tuned logic. That said, the hard part isn’t demos, it’s proving long-tail safety, debuggability, and regulatory trust when the decision-making is largely opaque. The gap between a great model and a deployable driver is still enormous.
This is a useful framing. Treating AI as an async collaborator instead of an instant oracle feels more realistic, especially as reasoning modes get slower. Good workflows should assume latency and help humans stay productive while the model is thinking, not just show a spinner.
Gumroad isn’t winning on features, it’s winning on boring infrastructure. Payments, taxes, refunds, and trust are the real moat, and most alternatives underestimate how hard those are to get right.
This post highlights Enshittification, which in genera; is the process where platforms start out serving users, then shift to exploiting users to benefit business customers, and finally hollow everything out to extract maximum profit for themselves.
This resonates. A lot of real technical talent starts with curiosity and boundary-pushing as a teenager. Treating that impulse as purely criminal instead of something to mentor and redirect feels like a great way to waste the next generation of engineers and security researchers.
Nice explanation. This is one of those shader tricks that feels obvious once you see it, but saves a ton of asset wrangling in practice. Treating black as a blend signal at render time instead of baking alpha channels is a clean solution, especially for effects, legacy textures, or quick prototyping.
This is the part that doesn’t get enough attention. The real risk isn’t any single vendor, it’s the aggregation layer. Once ALPR, retail cams, traffic cams, ISP data, and vehicle telematics all land in one searchable system, the idea that this will be perfectly RBAC’d and jurisdictionally contained is fantasy. At that point it’s not policing tech, it’s a nationwide surveillance substrate held together by policy promises.
the tool itself doesn’t change anything from the ISP’s perspective. It just fetches metadata and opens magnet links. What matters is what you download, where you live, and how your torrent client behaves, not whether you clicked the magnet in a browser or a terminal.
What’s interesting here is that Claude isn’t acting like a utility-maximizing agent at all, it’s behaving like a policy trained to model cooperative human norms. In repeated Prisoner’s Dilemma setups, that looks like tit-for-tat or conditional cooperation, but it’s really an artefact of training and prompt framing rather than strategic reasoning. It’s a good reminder that LLMs in game theory experiments are sampling from learned social distributions, not solving the game in its classical sense.
This resonates. Auth keeps getting more “secure,” but the failure modes are brutal for real users. We’re great at keeping attackers out and terrible at helping legitimate users recover when something goes wrong.
The interesting part here isn’t the PR angle, it’s what orbital compute actually enables. Satellites generate massive raw data streams that are bottlenecked by downlink bandwidth, so pushing inference and preprocessing into orbit changes the economics completely. You move from “send everything to Earth and hope for capacity” to “filter, compress, classify, and prioritize in situ.”
If SpaceX starts embedding real compute (GPUs, NPUs, or specialized ASICs) into Starlink nodes, you effectively get a globally distributed edge network with zero terrestrial dependencies. Blue Origin’s approach is more traditional and modular payload bays with dedicated compute racks, but the principles are the same: reduce latency for space-to-space communication and eliminate ground-station choke points.
The military and earth-observation implications are huge. Whoever controls orbital compute ends up owning the pre-processing layer for weather, climate, agriculture, ISR, maritime, and comms. This is less about “data centers in space” and more about creating an entirely new tier of cloud architecture that never touches the ground.
Amazing that a 10-day hack from 1995 is still quietly deciding what half the files on the Internet actually are. Meanwhile my “temporary” scripts from last week can’t survive a single deploy. The Web really is held together by archaeological layers of glue code nobody dares touch because it somehow… still works.
Starlink hitting vending machines is wild. That’s when you know the product has crossed from “early adopter tech” to “impulse buy on your way out of the store.” Kind of amazing that satellite internet has become plug-and-play enough for this to even make sense.
This is great work. Once ALPR coverage is dense enough that you can’t go anywhere without generating a permanent record, the “mere observation” argument falls apart. Mapping this openly is one of the few ways communities can actually understand what they’ve signed up for.
So now we’ve reached the point where “detox” means running your blood through a Brita filter? At this rate the wellness industry is going to reinvent dialysis and charge extra if you want the microplastics returned for recycling. Hard to imagine this doing much besides making your wallet noticeably lighter.
This resonates a lot. As founders we make way more emotional decisions than we admit, and half the battle is just getting out of our own way. The “AI CEO” framing is clever because it basically forces you into a discipline loop,focus on essentials, talk to users, cut the noise, execute consistently. It’s not that the AI is brilliant, it’s that it doesn’t get tired, anxious, or distracted like we do.
Sometimes that neutrality is exactly what a struggling startup needs.
This actually looks handy for the “small team with a couple of env files” use case. Most secret-management tools are great once you’re at scale, but trying to explain sops or git-crypt to a team that just wants to stop pasting secrets into Slack is… not fun. A simple password-protected vault committed to git is a reasonable middle ground.
I like the OS keyring integration too,removes a lot of friction. Curious how it behaves in multi-machine workflows and whether you plan to add any guardrails around accidental plaintext commits, since that’s usually where lightweight tools get tripped up.
At this rate Netflix isn’t building a streaming service, it’s building a monopoly starter pack. Give it a few more acquisitions and the “Are you still watching?” prompt will legally qualify as a government notice.
I was flabbergasted driving a 2023 Honda Civic Sport during a trip to Thailand, easily one of the best-balanced cars I’ve driven. Makes me wonder if that model is using the same engine they’re talking about here
This is fascinating work. Right now it’s mostly focused on near-death experiences, but I’d love to see it grow into a broader map of spirituality, consciousness studies, and related fields. The visual approach has a lot of potential once more domains are added.
So hair grows by pulling itself upward? Perfect. I’m just going to stand under a ceiling fan and wait for my hair to pull itself back onto my bald spots. Science has finally given me a plan.
Walrus isn’t trying to replace Kafka, but it does beat Kafka in a few narrow areas. It’s a lightweight Rust-based distributed log with a fast WAL engine and modern I/O (io_uring), so the operational overhead is much lower than running a full Kafka stack. If you just want a simple, fast log without JVM tuning, controllers, or the entire Kafka ecosystem, Walrus is a lot easier to run. Kafka still wins on ecosystem, connectors, and massive scale, but Walrus is appealing for teams that want the core idea without the complexity. Really impressed by the direction here, great work!!.
For this use case it makes sense. cJSON is a general-purpose library, but Redis only needs fast access to a few top-level fields for filters. A tiny purpose-built parser that avoids allocations and skips building a full tree will naturally beat a generic JSON parser here. It’s not “better than cJSON” in general, just better for this very narrow hot path.
I agree with you on SSDs, that was the last upgrade that felt like flipping the “modern computer” switch overnight. Everything since has been incremental unless you’re doing ML or high-end gaming.
When RAM gets so expensive that even Samsung won’t buy Samsung from Samsung, you know the market has officially entered comic mode. At this rate their next quarterly report is just going to be one division sending the other an IOU.
A lot of the time it’s less “nobody checked the security inbox” and more “the one person who understands that part of the system is juggling twelve other fires.” Security fixes are often a one-hour patch wrapped in two weeks of internal routing, approvals, and “who even owns this code?” archaeology. Holiday schedules and spam filters don’t help, but organizational entropy is usually the real culprit.
I get the frustration, but I think the npm option actually makes sense here. A lot of users who’d benefit from a fast Rust tool aren’t Rust developers and won’t have Cargo installed. Shipping it through npm lowers the barrier while still giving everyone the performance benefits. It’s not a knock on Cargo, just a way to make the tool more accessible.