Samuel Chazy

Head of AI Engineering | Context & Prompt Engineering | LLM, Generative, & Agentic AI | Microsoft Azure AI Certified | MSc in Artificial Intelligence | UAE

[Samuel Chazy @samuelchazy]

Designing agentic AI systems requires more than a powerful model. As AI moves to production, architecture matters. Clear role separation enables scalability, reliability, and governance. It is about clarity, not smarter models.

#AgenticAI #AIArchitecture #LLMEngineering

[Samuel Chazy @samuelchazy]

Negotiation in Agentic AI enables multi-agent systems to resolve conflicts, share resources & align decisions without centralized control. It’s structured autonomy in action, powering scalable coordination in complex systems.
#AgenticAI #MultiAgentSystems #AIArchitecture

[Samuel Chazy @samuelchazy]

Agentic AI isn’t just autonomy, it’s safe autonomy: acting independently but within rules & oversight. Built on self: organization, regulation, adaptation, optimization & determination, governance makes it scalable and trustworthy.

#AgenticAI #AIGovernance #ResponsibleAI

[Samuel Chazy @samuelchazy]

Defensive Prompt Engineering: stop prompt attacks such as Jailbreaking, Passive phishing, data leaks & injections. Solution: model-level defense, proposed by OpenAI = system prompt => user prompt => model outputs => tool outputs

#AI #Cybersecurity #PromptEngineering

[Samuel Chazy @samuelchazy]

I love this quote: "The journey from 0 to 60 is easy, whereas progressing from 60 to 100 becomes exceedingly challenging.". Ding et al. UltraChat. Demos are quick. Production-ready projects take time, polish, and persistence. That’s the real challenge.

#LLM #AgenticAI #AI

[Samuel Chazy @samuelchazy]

I just published OpenAI GPT-OSS Structure & Experimentation https://t.co/B9g00vXsq5

#OpenAI #ArtificialIntelligence #MachineLearning #OpenSourceAI #ReasoningModels #ChainOfThought #AIResearch #Innovation

[Samuel Chazy @samuelchazy]

LLMs can learn new patterns without fine-tuning via prompting, which acts as a low-rank weight update, similar to LoRA adapters. The image shows how Δ validation loss (in context learning) closely follows standard validation loss (fine-tuning). https://t.co/zbAVg17JJP

#LLM #AI

[Samuel Chazy @samuelchazy]

This is just amazing! Generating UI on the fly. GenUI.

[Google DeepMind @GoogleDeepMind]

Here's how Gemini 2.5 Flash-Lite writes the code for a UI and its contents based solely on the context of what appears in the previous screen - all in the time it takes to click a button. 💻 ↓

[Samuel Chazy @samuelchazy]

I came across this paper on Prompt Engineering. ProRefine, a teacher-student framework using LLM Task, Feedback, & Optimizer. Breaks tasks into smaller steps & refines them in an iterative way, perfect for agentic AI workflows. https://t.co/rGwwRbVsJj

#PromptEngineering #LLM #AI

[Samuel Chazy @samuelchazy]

Here’s a fun GenAI video I created. Prompt: ChatGPT - Original AI Image: Adobe Firefly - AI Animation: Runway - AI Music: Suno - Final cut: Premiere Pro. When AI tools combine, the possibilities are endless.
#GenAI #AIvideo #ChatGPT #RunwayML #SunoAI #AdobeFirefly

[Samuel Chazy @samuelchazy]

There's excitement in the AI community about Model Context Protocol (MCP), by Anthropic. It simplifies AI-powered tools integration with data sources. This video demonstrates how MCP tools enable fast, flexible web scraping. #MCP #AI #Automation #Productivity #Anthropic

[Samuel Chazy @samuelchazy]

AI agents are transforming industries! Imagine a smart assistant that finds your perfect home, streamlines travel, or handles customer inquiries: all through natural conversation. The future of automation is here.

#AIInnovation #AIagents #Automation #FutureOfWork

[Samuel Chazy @samuelchazy]

2025 is set to be the year of #AIVoice! We're shifting from dull text chats to dynamic voice interactions that capture sentiment & tone. But with real-time responses comes the need for robust guardrails to keep our systems secure. #Innovation #AI #VoiceTechnology #TechTrends

[OpenAI @OpenAI]

Trading Inference-Time Compute for Adversarial Robustness https://t.co/nDZYVMSAh7

[Samuel Chazy @samuelchazy]

RAG enhances data augmentation but faces latency issues. Optimize with query classification module using HyDE for retrieval, TILDE for reranking, Reverse for repacking, & Recomp for summarization to achieve the highest average score. Paper: https://t.co/hJGp9UaG9o

#LLM #AI #NLP

[Samuel Chazy @samuelchazy]

Tired of dry textbooks? An engaging way of listening to papers - Notebook by Google allows anyone to upload a document & turn it to a friendly conversational podcast between 2 people. A great educational resource.

https://t.co/GyYQdoYFDF
#LLM #AI

[Samuel Chazy @samuelchazy]

Superposition prompts are DAGs. Nodes are word sequences & edges are relationships. A words-sequence attends to words in other words-sequences if & only if there is a path. This concept resembles a graph network but works in a different way. https://t.co/XPqV9bTSOK
#LLMs #prompt

[Samuel Chazy @samuelchazy]

While LLMs have made advances in natural language processing, they face challenges with other languages, handling conjugation, diacritics, and pronouns. Integrating Knowledge Graphs with LLMs, makes Arabic chatbots produce a better and a more context-aware responses.

#LLM

[Samuel Chazy @samuelchazy]

على الرغم من التقدم الكبير الذي أحرزته النماذج اللغوية الكبيرة في معالجة اللغة الطبيعية، لكنها لا تزال تواجه تحديات مع لغات مثل اللغة العربية، خاصة فيما يتعلق بالتعامل مع التصريف، التشكيل، والضمائر. دمج الرسوم البيانية المعرفية مع النماذج اللغوية الكبيرة، تنتج إجابات أفضل.

#LLM

[Samuel Chazy @samuelchazy]

Experimenting with Knowledge Graphs & LLMs shows promising results, but challenges remain, esp. with context, semantics, and non-English texts such as Arabic. Combining RAG + KGs improves LLM accuracy. Check out this paper on LLMs & KG: https://t.co/NU8QGmjqd5

#LLMs #KG