Gokul Dhamodaran

Most slept on repo of all time, anyone can autoresearch & deeply recursively optimise anything. Do yourself a favor, get a cheap llm sub, or run something locally on some hardware, give it autoresearch and target it at something in your life. - Budgeting - Performance optimisations - Resume & job application machine - Local city/town mapping - Clear all your personal info off the web - Rebuild everything you pay for I've been doing this for months, my life is infinitely better. The concept is very very simple. 1. "Program.md": explain what it is you want, and what isn't allowed during autoresearch, especially great if u can define the output at each turn 2. "Prepare . py": a script that cleans up and enforces the experiment structure 3. Target: whatever it is you're trying to optimise, whether that's your email inbox, some mobile app/website you're trying to optimise load times for, or an llm training run. 4. Document and learn from the runs, this is practically free progress, if the run is designed as a ralph loop where the model is freshly instantiated at every turn even Qwen3.6-27B will work perfectly (heard from very trusted individuals.) This can revolutionise so many industries and lives, I fully believe this with all my heart.

Fleetbase is an open-source modular logistics operating system that enables developers to build custom supply chain solutions, manage deliveries, optimize routes, and track orders using a comprehensive API. github.com/fleetbase/flee…

Dr. Roger Seheult is a critical care physician who says sunlight does something to your body that no vitamin D pill can replicate. He just revealed 10 effects of sun exposure most people have never heard of (THREAD): 1) Infrared light penetrates up to 8cm into your body

Introducing Applied Electrodynamics. What would the world look like if we could see beyond visible light, across the entire electromagnetic spectrum? The way waves interact with matter, at every wavelength, encodes far more than just shape and color. It reveals internal structure, temperature, chemical composition, and more: a whole layer of the physical world that's invisible to us today. That's what we're building at Applied Electrodynamics: the infrastructure to fully perceive the physical world. WaveSight is the first piece of the puzzle. It brings radio frequencies into human perception, revealing the internal structure of objects behind opaque layers. Designed for construction crews, industrial inspectors, and security personnel. See through walls: ae-dyn.com

1/ We are building a new model system, where the harness, context, memory, and action space are separate from any one model family, thereby moving the frontier on the cost -to-outcome curve. And it’s not just about cost. It also has the added benefit of business continuity and resilience because every model is substitutable. This is the system we are using in our products, with great results. And we are making it available for our customers via Foundry too. microsoft.ai/news/optimizin…

infisical is prob one of the most underrated startups rn building for the agent native world

Cursor changed how people write code. Someone just built the same thing for Minecraft. It’s called Steve. Press K, type what you want, “mine some iron,” “build a house,” and an AI agent actually plays the game, navigating, mining, and building block by block, not just answering questions about the game. → Say “mine some iron” and the agent reasons about depth, navigates to it, locates ore veins, and extracts the resources itself, no scripted macro, actual reasoning per request → Ask for a house and it considers available materials, generates a structure, and builds it block by block → Multiple Steves can build together using deterministic spatial partitioning, each agent claims a section of the structure atomically, so they don’t collide or duplicate work → Powered by LLMs like OpenAI, Groq, and Gemini, plugged in through config → Each Steve is a real custom entity registered through Forge’s system, extending vanilla pathfinding so it moves through the world like any other mob, not a floating camera hack → Memory management prunes old context while keeping the last 10 exchanges plus current position, inventory, and known world features, so long play sessions don’t blow up the context window Still early: no crafting yet, actions run synchronously, and memory resets between sessions, per the project’s own roadmap. Open source, MIT licensed.

YOUR AI AGENT IS WASTING 61% OF ITS TOKENS ON REDUNDANT CONTEXT. TENCENT JUST OPEN-SOURCED THE MEMORY SYSTEM THAT FIXES IT AND PROVED IT ON SWE-BENCH, WIDESEARCH, AA-LCR, AND PERSONAMEM. If you're building AI agents, and you've been trying to solve long-term memory by dumping everything into a flat vector store and hoping the retrieval finds the right chunk, this changes the entire approach. It's called TencentDB Agent Memory. A 4-tier progressive memory pipeline that runs 100% locally with zero external API dependencies, MIT licensed, integrated with OpenClaw and Hermes agents in one command. Here's what the numbers look like on real benchmarks: → WideSearch -- token usage -61.38%, task pass rate +51.52% (33% → 50%) → SWE-bench -- token usage -33.09%, pass rate +9.93% (58.4% → 64.2%) → AA-LCR -- token usage -30.98%, pass rate +7.95% → PersonaMem -- accuracy jumped from 48% to 76% (+59%) These are measured on continuous long-horizon sessions -- SWE-bench runs 50 consecutive tasks in a single session to simulate real-world context-accumulation pressure. Not cherry-picked short-window comparisons. Here's the actual architecture: Long-term memory is a semantic pyramid, not a flat blob: → L0 Conversation -- raw dialogue → L1 Atom -- atomic facts → L2 Scenario -- scene blocks → L3 Persona -- user profile The agent normally reads the top layer (Persona). When it needs details, it drills down to Scenarios, then Atoms, then original conversations. Every level traces back to ground-truth evidence with no lossy compression. Short-term memory uses symbolic Mermaid canvases: → Verbose tool logs (hundreds of thousands of tokens) get offloaded to external files → A lightweight Mermaid canvas with node_ids stays in context (a few hundred tokens) → Agent reasons over the symbol graph, drills down by node_id when it needs the raw text Here's the wildest part: It's fully white-box debuggable. Every intermediate layer is a readable Markdown file. L3 Persona lives in persona.md. L2 Scenarios are Markdown blocks. Short-term canvases are Mermaid you can render in any editor. When recall goes wrong, you open the files and walk the chain "Persona → Scenario → Atom → Conversation" until you find the root cause. Most memory systems are opaque vector databases where you get a similarity score and no way to debug. This one you can literally cat the files. Deployment: → OpenClaw plugin: 2 commands, zero config, defaults to local SQLite + sqlite-vec → Hermes agent: Docker image with everything bundled, or attach to existing install → Hybrid retrieval built in: BM25 + vector + Reciprocal Rank Fusion → Optional Gateway API key + CORS allow-list for production → Works with any OpenAI-compatible LLM endpoint 9K GitHub stars. 836 forks. On Trendshift. TypeScript 84%. MIT license. 100% open source. (link in the comments)

Software quality now depends on the constraints you set around your agents. When humans manually wrote most of the code we could look at the code itself for signs of quality. Is it clean? Is it thoughtful? Is it fast? Can another engineer understand it? Does it have tests? Agents can now generate more code than people can read. When code generation scales beyond review, quality - checks for one or more of correctness, maintainability, security, performance etc - increasingly has to live somewhere else. It moves into the harness, environment and operating system around the agent. This can be the tests and deterministic checks that decide what the system is allowed to do (amongst others). Your constraints are what may eventually enable loops of agents to deliver production software reliably. They can include unit tests, property tests, acceptance tests, mutation testing and quality metrics. This back-pressure lets the system resist bad work before it becomes somebody elses problem. Set your constraints. They decide whether the code your agents generate is good enough to ship.

This tool is literally Higgsfield AI but FREE for good. It's called Wan2GP. A full AI video studio built specifically for people without expensive hardware. Runs on as little as 6GB of VRAM, even old RTX 10-series cards and 8GB laptops. Everything stays on your machine, no uploads, no caps, no watermarks. What you get in one app: • Text-to-video and image-to-video generation • The best open models built in: Wan 2.2, LTX-2, Hunyuan Video, Flux • A full browser interface with a queue system • LoRA support to customize any model • Mask editor and prompt enhancer included A 5-second clip generates in minutes on a mid-range gaming rig. No subscription, ever. 100% Free. Open Source.

One of the biggest divides in society today is cognitive. The top 1% protect their attention, block out focus time, and execute long-term targets. The rest of the world is trapped in a cycle of algorithm-driven dopamine.

We’ve gone from AI defaulting to designs that look like purple slop to those that have that recognizable Claude look. To avoid this, I always create a design md file with guidelines for color, typography, and more. Here are a few places to find inspiration for your design md: 1. Go to @mobbin and install its MCP to find design patterns and app screens you like. Paste the screens and ask AI to make a design md. 2. Check out designmd sh, which has a library of design md files for popular websites and apps. 📌 I walk through an example of creating a design md in my new tutorial: youtube.com/watch?v=G9o8eo…

You'd bounce back from failure faster than 99% of people if you followed this EXACT framework tonight:

unsurprisingly the website this is from is literally cookingforengineers dot com

兄弟们，微软居然还搞了这个好东西，值得看看啊！ 做零售店、巡检的行业人简直是大喜~ Microsoft扔出一个仅仅4B的流式视觉模型，能实时盯着视频说“什么时候发生了什么”。 Mage-VL是Codec-Native的Streaming VLM，专门做直播式理解。 它一边看视频一边描述正在发生的事，你还能用自然语言指定关心的时刻，比如“火车到站的时候”“进球的瞬间”。4B体量却能持续处理流式输入，效率很高。 以前实时视频理解要么靠大模型烧算力，要么只能事后分析。

I wish I knew this 7years ago

text-to-cad just crossed 10,000 stars you can use it to generate: - STEP files - URDF, SDF and other sim artifacts - 3D mesh files like STL, 3MF and GLB - gcode for 3D printing - DFM checks for popular services like sendcutsend 100% free and open source ✌️

@Mechramc @karpathy @rasbt Wishing you the best !!

At 38, I had 17 years of program management under my belt. PgMP, PMP, enterprise SaaS. Zero machine learning experience. At that time, as everyone does, I started doing some Coursera courses but everything was vague. There were two things that helped - @karpathy and @rasbt. Once the concepts were broken down - I could build, break, and eventually come up with 100s of ideas that failed. I decided to pivot into AI research. Not a side hobby. A real pivot. I spend to this day at least a few hours every week reading latest papers and trying to break the implementation of the paper. Even now, I am participating in the ICLR huggingface paper reproduction challenge. Two years later at 40: 6 papers on arXiv, an NVIDIA Inception membership, founded Murai Labs, and an open source Tamil language model that is coming out soon. The book grew directly out of what I needed to build TamilLM from scratch. If it helped me learn how to ship a real model with no ML background, it can help you too. It is such a beautiful coincidence that TamilLM finished pretraining and Under The Hood launched on Amazon on the same day - it was not planned. (Kadavul irukaan Kumaru moment) 100+ copies sold on Leanpub. #1 bestseller there. Now available worldwide on Amazon. 8 free chapters at murailabs.com/under-the-hood (Chapters 4, 5, 8, 13, 18, 23, 27, 32). India Kindle: amazon.in/dp/B0H9FYVK3W US Kindle: amazon.com/dp/B0H9FYVK3W US Paperback: amazon.com/dp/B0H9FWPCDV Leanpub: leanpub.com/under-the-hood I started late. But not done yet - I am glad I started. 🙏

Introducing Waddle Labs: Claude Code for robots. Connect our API to your robot and enter a prompt, then our agents write code to achieve the task in 20 minutes. @yiding_song @theWaddleLabs

We’re open-sourcing PorTAL, our framework for shared task representations and cross model LoRA adaptation. It now spans from hybrid attention models to multimodal systems including Gemma 4 E2B, Mistral 7B & @thinkymachines' Inkling. Code: ramp-public/portallib Models: @huggingface /RampPublic