Dinesh Kumar Subramanian

I tweet about AI, design, full-stack development, and cybersecurity.
CTO @finsteinA

[Dinesh Kumar Subramanian @ortist]

@mynameisyahia @getcontextdev Working on a agentic master data management platform would like to have access to https://t.co/fOOm59O3N5 and i have already signed up

[Nadieh Bremer @NadiehBremer]

📣 NEW! I’ve just released the BIGGEST and perhaps most creative project I’ve ever worked on!

“Searching for Birds” https://t.co/CgFoywkrjq 🐤

A #dataviz article & exploration that dives into the data that connects humans with birds, by looking at how we search for birds.

[Alex Vacca @itsalexvacca]

CIA can't operate without it.
Pentagon can't function without it.
And Wall Street can't trade without it.

Yet most people have no idea about what Palantir does.

How the Government let a $300 Billion surveillance company track you everywhere 🧵

[Aurimas Griciūnas @Aurimas_Gr]

Adding 𝗠𝗖𝗣 to 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗔𝗚 Systems.

If you are building RAG systems and packing many data sources for retrieval, most likely there is some agency present at least at the data source selection for retrieval stage.

This is how MCP enriches the evolution of your Agentic RAG systems in such case (𝘱𝘰𝘪𝘯𝘵 2.):

𝟭. Analysis of the user query: we pass the original user query to a LLM based Agent for analysis. This is where:

➡️ The original query can be rewritten, sometimes multiple times to create either a single or multiple queries to be passed down the pipeline.
➡️ The agent decides if additional data sources are required to answer the query.

𝟮. If additional data is required, the Retrieval step is triggered. We could tap into variety of data types, few examples:

➡️ Real time user data.
➡️ Internal documents that a user might be interested in.
➡️ Data available on the web.
➡️ …

𝗧𝗵𝗶𝘀 𝗶𝘀 𝘄𝗵𝗲𝗿𝗲 𝗠𝗖𝗣 𝗰𝗼𝗺𝗲𝘀 𝗶𝗻:

✅ Each data domain can manage their own MCP Servers. Exposing specific rules of how the data should be used.
✅ Security and compliance can be ensured on the Servel level for each domain.
✅ New data domains can be easily added to the MCP server pool in a standardised way with no Agent rewrite needed enabling decoupled evolution of the system in terms of 𝗣𝗿𝗼𝗰𝗲𝗱𝘂𝗿𝗮𝗹, 𝗘𝗽𝗶𝘀𝗼𝗱𝗶𝗰 𝗮𝗻𝗱 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗠𝗲𝗺𝗼𝗿𝘆.
✅ Platform builders can expose their data in a standardised way to external consumers. Enabling easy access to data on the web.
✅ AI Engineers can continue to focus on the topology of the Agent.

𝟯. Retrieved data is consolidated and Reranked by a more powerful model compared to regular embedder. Data points are significantly narrowed down.
𝟰. If there is no need for additional data, we try to compose the answer (or multiple answers or a set of actions) straight via an LLM.
𝟱. The answer gets analyzed, summarized and evaluated for correctness and relevance:

➡️ If the Agent decides that the answer is good enough, it gets returned to the user.
➡️ If the Agent decides that the answer needs improvement, we try to rewrite the user query and repeat the generation loop.

Are you using MCP in your Agentic RAG systems? Let me know about your experience in the comment section 👇

#LLM #AI #MachineLearning

[Jon Willbanks @jonwillbanks]

Try my 30-day challenge:

12oz of this ORS first thing in the morning and before bed.

You'll likely notice better energy, fewer cravings, deeper sleep, and less nighttime peeing.

Your brain and mitochondria will thank you.

[Jonathan Jacobi @j0nathanj]

Introducing Multiverse: the first AI-generated multiplayer game.

Multiplayer was the missing piece in AI-generated worlds — now it’s here. Players can interact and shape a shared AI-simulated world, in real-time.

Training and research cost < $1.5K. Run it on your own PC.

We @enigma_ai are open-sourcing everything: code, data, weights, architecture, and research.

Here’s a closer look (and how it works!):

[Jie Liu @jie_liu1]

After hacking GPT-4o's frontend, I made amazing discoveries:
💡The line-by-line image generation effect users see is just a browser-side animation (pure frontend trick)
🔦OpenAI's server sends only 5 intermediate images per generation, captured at different stages
🎾Patch size=8

[Shivers @thinkingshivers]

With 4o Image Generation, it's time to revisit Scott Alexander's bet against Vitor. Although Scott declared victory back in Sep. 2022, some commenters (myself included) felt this was dubious.

For each prompt, I'll generate exactly four pictures and post all of them.🧵

[Min Choi @minchoi]

This is wild.

Vibe coding with AI just completely changed the game.

People can't stop creating games with Grok, Windsurf, Cursor &amp; Claude 3.7 Sonnet.

10 wild examples:

1. Fortnite meets Minecraft on Three.js 🤯

[Philipp Schmid @_philschmid]

Here is your Weekend Project! 🚀 @UnslothAI and @huggingface released an example notebook on how to make @GoogleDeepMind Gemma 3 think using RL with GRPO in a free @GoogleColab Notebook!

https://t.co/oCM1OQKUd8

[anslo @anslogen]

Introducing Slow Roads 2.0 - procedurally-generated, scenic roads for endless driving zen. Built with
@threejs to run in your browser - no installs, no ads, just roads.

Play free at https://t.co/8FtBldzqKG, and join me on Bluesky for future updates

#threejs #webgl #procedural

[Roy Carrilho @RuiCarrilho5]

Found Let's Build a Simple Database! A nice little resource that teaches you how to build a database from scratch in C. Enjoy!

[Indie Hackers @IndieHackers]

Vibe Coding for Dummies 📗

We collected the best tips on X for vibe coding, so you can build cool shit without getting stuck in a loop (bookmark this):

[Unwind AI @unwind_ai_]

Opensource AI Agent that can use Computers and Smartphones fully autonomously.

Agent S2 outperforms OpenAI Operator and Anthropic Computer Use with Claude 3.7 Sonnet.

It plans its actions, prevents errors proactively and learns from past experiences.

100% opensource and free.

[Cristian Peñas ░░░░░░░░ @ilumine_ai]

This quick experiment I just did made my jaw drop...

You can literally create and play any game by iterating over images with the new Gemini model! 🤯

[Nicolas Zullo @NicolasZu]

It's out!! The ULTIMATE guide to vibe coding games.

How did I do it? 20 hours. 500 prompts. 20 euros. That's all it took to make a multiplayer 3D game 100% with AI, 0 human code, not even a small edit. 1.5M views on X. 45k tested the game. Crazy!

I have received hundreds of messages asking me how I did it. People even offered (big) money for my prompts. It’s just insane.

Tonight I share in detail my secret sauce, so you can start yourself in 5 min! As of now, your limit = your imagination.

For context, last week, @levelsio started the trend to make an entire game just by prompting it. A concept coined as "vibe coding". I decided to take upon the challenge and humbly try to beat Pieter in terms of look and feel, to see if it was in my reach, and knowing I've been using Cursor for 12 months now (> 3000+ prompts)

Get the full guide in comment, and try the game (link also in next posts) and judge for yourself. Fly around the clouds and think about the fact the game was 100% made with AI. Unreal.

And now, YOU can vibe code a game. Enjoy!!

[Kieran Klaassen @kieranklaassen]

Stop vibe coding, just use the app immediately.

Experiment: This flow creates real-time app experiences from simple prompts, generating interactive screens that adapt to your responses.

[Yichao 'Peak' Ji @peakji]

Hi! I'm Peak from Manus AI. Actually, it's not that complicated - the sandbox is directly accessible to each user (see screenshot for method).

Specifically:
* Each session has its own sandbox, completely isolated from other sessions. Users can enter the sandbox directly through Manus's interface.
* The code in the sandbox is only used to receive commands from agents, so it's only lightly obfuscated.
* The tools design isn't a secret - Manus agent's action space design isn't significantly different from common academic approaches. And due to the RAG mechanism, the tools descriptions you get through jailbreaking will vary across different tasks.
* Multi-agent implementation is one of Manus's key features. When messaging with Manus, you only communicate with the executor agent, which itself doesn't know the details of knowledge, planner, or other agents. This really helps to control context length. And that's why prompts obtained through jailbreaking are mostly hallucinations.
* We did use @browser_use's open-source code. In fact, we use many different open-source technologies, which is why I specifically mentioned in the launch video that Manus wouldn't exist without the open-source community. We'll have a series of acknowledgments and collaborations coming up.
* There's no need to rush - our team has always had an open-source tradition, and I personally have been sharing my post-trained models on HuggingFace. We'll be open-sourcing quite a few good things in the near future.

[Okara @askOkara]

I spent hours analyzing v0’s 2000+ line system prompt.

It's one of the best examples of prompt engineering I've seen

Let me break it down 🧵

[Victor M @victormustar]

QwQ-32B changed local AI coding forever 🤯

We now have SOTA performance at home. Sharing my stack + tips ⬇️