Sami Rm.

AI Engineer & Researcher | Agentic AI • Computer Vision | TypeScript • Python • AWS

[Sami Rm. @__samiur]

Here's a 30-second tour of Proloquent, the LanguageCert prep app I've been building. It covers full mock exams, instant feedback, and a live AI speaking examiner.

Try it free at https://t.co/ha1R45Tk24, and I'd love to hear what you think.

[Proloquent @proloquent]

AI-powered LanguageCert prep 🇬🇧

All 4 skills, instant scores, timed mocks, and speaking with a live AI examiner.

Start free 👉 https://t.co/T3hUX0EsEy

#LanguageCert #AISpeaking #EnglishTest

[Sami Rm. @__samiur]

This is the trap: the US won't slow down because China might not, and China won't because the US might not. Everyone assumes caution means losing, and nobody ever checks if that's true

[Jacob Coxon @hilbertspaess]

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

[Sami Rm. @__samiur]

Proud to share what we've been building at Material Index.

Over the past months I've worked on these products end to end - AI-enabled audits that catalogue an asset from a single photo, pathway insights that surface reuse and recycling routes, our MCP integration with Claude, and MI Chat, a project-aware assistant that answers in the language of construction reuse.

A few things I learned along the way:

- Getting AI to map a photo to the right Uniclass category is much harder than getting it to describe the photo.
- The useful part is not automation on its own, it is keeping the auditor in the loop while cutting the repetitive work.
- Real value shows up when the data flows straight into the reports teams already rely on.

Now live on real projects with Derwent London and British Land. More to come

[Sami Rm. @__samiur]

'I don't know' is the most accurate AI prediction and the least retweetable one

[Ravid Shwartz Ziv @ziv_ravid]

The AI hype about replacing all of us has accelerated massively in the last few weeks.

But the main problem with AI predictions is that Nobody actually knows what AI will achieve.

Will the small bugs preventing end-to-end pipelines get solved? Will scaling keep working? No one knows.

But on Twitter, you get two types of people talking about it:

- People with very strong opinions who have no idea about the actual capabilities of future models. They just talk with confidence.

- And CEOs of OpenAI/Anthropic/xAI who do know more, but have so many incentives that it's hard to tell what's real and what's PR. (What about GPT-2 weights? Are they still dangerous?)

The honest answer is "I don't know." But that doesn't get engagement....

[Sami Rm. @__samiur]

@TechWithTimm Absolutely. And while we're at it, it'll also understand why that one function from 2019 exists and why nobody's allowed to touch it 😅

[Sami Rm. @__samiur]

The real win wasn’t the architecture. It was making AI fit how auditors actually work.

#AIEngineering #LLM #CircularEconomy #Sustainability #SoftwareArchitecture #AppliedAI #SystemDesign #HumanInTheLoop

[Sami Rm. @__samiur]

Last December, we shipped a new AI feature that made building audits 3x faster - without hiring more people.

It lets auditors upload images, walk away, and come back to analysed assets

[Sami Rm. @__samiur]

Under the hood: Multi-agent AI workflows, Queue-based processing, Parallel Lambda workers, and Vector search with LLM re-ranking for accuracy.

[Sami Rm. @__samiur]

So we flipped the workflow: people move on, AI works in the background.

- Upload everything in a single action

- AI works independently on each asset in the background

- Processing scales automatically behind the scenes

- Humans remain the final decision-makers

[Sami Rm. @__samiur]

At Material Index, Our auditors used to photograph hundreds of construction assets per audit - then manually classify each one: title, category, dimensions, materials, quantities and more. It was slow, boring and repetitive.

[Sami Rm. @__samiur]

(4/4)

Right now, it works decently with the YOLO-World model, but fine-tuning YOLO8L or YOLO11L on aerial data is the next step to boost accuracy and make it deployment-ready.

Tech Stack:
FastAPI • YOLO • OpenCV • OpenAI • LangGraph

#MultiAgentAI #CV #YOLO #LangGraph

[Sami Rm. @__samiur]

What if flying taxis could "see" danger before it happens?

This weekend I experimented with SkyLaneAI - a real-time hazard detection system for aerial vehicles.

Here's how it works 🧵

[Sami Rm. @__samiur]

(3/4)

Why Multi-Agent AI?

Instead of one AI doing everything, I built specialized agents:

-> Detection Agent → Finds hazards
-> Context Agent → Analyzes patterns
-> Alert Agent → Generates warnings
-> Message Agent → Creates clear alerts

[Sami Rm. @__samiur]

54+ comprehensive guides covering:
📚 ML algorithms from scratch
🧠 Deep Learning (PyTorch/TF/JAX)
👁️ Computer Vision (YOLO → Diffusion Models)
💬 NLP (BERT, GPT, Transformers)

Complete learning path from fundamentals to cutting-edge AI

[Sami Rm. @__samiur]

I'm excited to share my comprehensive educational repository covering AI, ML, and Data Science - from fundamentals to advanced AI topics.

🔗 Check it out: https://t.co/M9Fu4WGODj

[Sami Rm. @__samiur]

Try tough queries (“Predict next visible ISS pass?”), explore the code, and PRs welcome:
https://t.co/7jQKho01QP

#Nextjs #VercelAISDK #TypeScript #Zod #NASA #SpaceX #LaunchLibrary #AIAgents #ModelContextProtocol

[Sami Rm. @__samiur]

Shipping Galactic Grid — an AI-powered space dashboard with a full Model Context Protocol (MCP) implementation. Ask “Where’s the ISS right now?” and the assistant calls real tools to fetch live data + stream answers.

Live: https://t.co/Wuw4oAvZmy

#MCP #AI #SpaceTech #OpenSource