The routing is done finally! Now DRC and fine tuning! 🚀⚡

It is getting interesting 🤔

Özdisan sözde Cumhuriyet Bayramını kutluyor 😀 Resimlere yakından bakınca yüzler bir ilginç duruyor. AI olmasın sakın.

Hayattaki mottom. Hiç bir şeye başlamaktan korkmadım. Bazen çok hesap kitap yapmak önünüze engel koyuyor. Ummadığınız yerden fırsat çıkıyor.

The hack is more professional than the product itself 😀

Herkesi böyle kabul edip API aldırmaya alıştırıyorlar. Bu tazaga ben de düştüm 😅

Love screens 😍 😀

Deepseek v4.1 Flash vs GLM 5.3 Flash on an embedded SW task. Deepseek wins here, faster and cheaper although it is more expensive per token. It does fewer calls, hence cheaper. However both couldnt solve the real simple issues.

Are we overestimating the long-term value of tools in the age of advanced LLMs? I've been thinking about something lately. Many startups today are building tools that essentially act as wrappers, bridges, or abstractions around complex workflows. But if LLMs continue improving at their current pace, does this model still make sense in the long run? If an LLM can: •Understand the problem •Generate the code •Build the workflow •Create the interface •Integrate systems •Explain how everything works ...then why wouldn't users simply ask the LLM to generate their own custom solution instead of buying a predefined tool? Of course, there will always be a place for products that provide trust, support, compliance, distribution, and great user experience. But many software categories seem increasingly vulnerable. Take software development as an example. IDEs are becoming AI-first. Framework complexity is being abstracted away. Platforms that once provided significant value through automation are finding that LLMs can now reproduce much of that value directly. The same question applies to hardware and electronics. For decades, we've relied on layers of tools to simplify development. But when an LLM can design schematics, review PCBs, write firmware, generate drivers, create test plans, and even explain tradeoffs, the value of many intermediate tools starts to look different. Perhaps the future is not "more tools." Perhaps the future is fewer tools, with increasingly capable AI acting as the universal interface between humans and technology. I don't think tools will disappear entirely. Rather, many of today's tools may evolve into infrastructure that AI agents consume behind the scenes, while human users interact primarily with the AI itself. Curious to hear what others think: Will AI largely replace software and hardware development tools, or will those tools simply evolve and become invisible layers underneath AI-driven workflows? #AI #LLM #ArtificialIntelligence #SoftwareDevelopment #EmbeddedSystems #HardwareEngineering #FutureOfWork #Automation #ProductStrategy #TechInnovation

This is my version of hello world on the screen

Ben de teyip falan parçalayıp dayak yiyordum

Mistakes happen, but I can fix it 💪🏼

This is how it will feel in couple of years with the AI. We pour immense resources into beta versions, proofs-of-concept, and endless MVPs, often without a scalable, long-term plan for maintenance, documentation (note that dusty, open manual!), or a clear decommissioning path. Are we so focused on shipping the next shiny object that we’re forgetting the foundation?

One SDK. Three power classes. The hard part of edge AI is rarely the model. It is that every silicon vendor ships its own toolchain, and switching parts means rewriting the application. E1M modules are pin-compatible across an open mechanical and electrical standard — two footprints, one contract. E1M-AEN at 0.5 TOPS for always-on sensing on Alif Semiconductor Ensemble, E1M-X V2N at 4 TOPS for real-time vision on @RenesasGlobal RZ/V2N, and E1M-X V2N-M1 at 4 + 25 TOPS with the DeepX DX-M1 when the model no longer fits. Same board.yaml. Same C/C++ API on top of Arm CMSIS. Zephyr, Yocto or bare-metal, depending on the target. Pinout and mechanical envelope are public in alplabai/e1m-spec. Evaluation kits: E1M EVK and E1M-X EVK. → alplab.ai @alplabai #edgeAI #embeddedsystems #SoM #ArmCortex #openhardware

Just received another investor rejection. The reason: a System-on-Module (SoM) is "not a scalable solution" for them. Here's the irony. Behind almost every complex product, there's a SoM. Many of the robots, drones, and smart devices investors love to fund run on one. But a robot is easy to picture. A drone is easy to picture. The module that makes them possible is much harder to grasp in a 10-minute pitch. What I'm learning: you don't get funded for what you build. You get funded for what investors can easily understand. Founders building deep tech and infrastructure, how do you bridge that gap? #DeepTech #Hardware #EmbeddedSystems #Startups #Fundraising

Steel does not pass a radio link. Inside a pipe there is no operator, no uplink and no second chance to look. Everything an inspection crawler decides, it decides with the compute it carried in. Recipe 11 runs corrosion and crack segmentation per frame on an E1M-X V2N — 4 TOPS, inside the battery budget. Classification happens during the run, so what comes back out is a list of flagged locations with odometry, not hours of video for somebody to watch afterwards. One board.yaml: A55 on Yocto for the vision pipeline, M33 on Zephyr for the drive loop. Same SDK, same C/C++ API. Come back with findings, not footage. → alplab.ai @alplabai #edgeAI #industrialinspection #embeddedvision #pipelineintegrity #NDT