aiXplain

ورا أي Benchmark قوي الـ Dataset اللي انبنى عليها! مشان نطلع بنتائج دقيقة ومحايدة لتقييم نماذج تحويل الصوت لنص (ASR)، كان الاعتماد على dataset مخصصة ومبنية من الصفر لهاد التقرير والأرقام بتحكي عن حالها: حجم الداتا: أكثر من 11 ساعة صوتية (2,340 مقطع). التنوع الإقليمي: 4 لهجات رئيسية (شامية، مصرية، حجازية، نجدية). تنوع واقعي: أخدنا مقاطع من بودكاست، أخبار، وحوارات ببيئات مختلفة (صوت نظيف، ضجة، موسيقى، أصوات بعيدة) ومن فئات عمرية مختلفة للذكور والإناث. بس النقطة الأهم هون؟ التفريغ الصوتي (Ground Truth) انعمل يدوياً 100% بواسطة مدققين بشر ، بدون أي تدخل أو مساعدة من نماذج AI. ليش هاد مهم؟ مشان نمنع أي انحياز (Bias) لنظام معين وتكون المنافسة عادلة وصافية. نتائج أي benchmark محكومة بالـ dataset تبعها. لو الداتا اللي عم تختبر عليها ما بتشبه الـ Production Use Case تبعك بالواقع.. ترتيب النماذج وأرقامها ما رح تعنيلك شي. التفاصيل كاملة وكيف تمت الاختبارات بتلاقوها بالتقرير الشامل هون: linkedin.com/posts/aixplain…

Our Chief Ethics Advisor Dr. @monamhamdy joined @MalakFouad on @WhatIDidNextPod for a conversation about AI policy. A thoughtful look at where the rules are heading and what responsible AI looks like in practice. ” We say things like, AI’s gonna get really good. All the jobs are going to be gone, and you’re going to get something like a universal basic income, some kind of allowance for being human. That’s a policy. That is not a framework or a vision or a philosophy about what we want this technology to do.” Watch the full episode here: youtube.com/watch?v=csCpmy…

Our CEO, Dr. @Hassan_Sawaf , on Access Middle East with @CNBC @dan_murphy , talking about how aiXplain’s Agentic OS empowers enterprises to move from AI experimentation to real business impact faster. Dr. @Hassan_Sawaf explains how @aixplain can be your technical partner to scale and adapt your business without being locked into specific providers or architectures, with seamless, enterprise-ready governance and control. Read full interview here: aixplain.com/blog/aixplain-…

Our Chief Ethics Advisor dr. @monamhamdy sat with @MalakFouad on @WhatIDidNextPod and talked all about the future of AI & its ethical implications. “All technology is built on algorithms that humans created while observing their natural world.” @monamhamdy Watch the full episode here: youtube.com/watch?v=csCpmy…

Our CEO Hassan Sawaf on Access Middle East with @CNBC's @dan_murphy , talking about AI orchestration and what it actually means for enterprise operations. Dr. @Hassan_Sawaf breaks down how @aixplain's AgenticOS acts as the connective fabric, bringing chips, cloud services, and the latest models into one governed platform. Built for the region’s AI Sovereignty, deployable , and designed so organizations can build their own agents without stitching it all together from scratch. Check the full interview here: aixplain.com/blog/aixplain-…

هل ياتُرى نماذج التعرّف الآلي للكلام تفهم اللهجة المصريّة ؟ معظم اختبارات نماذج التعرّف على الكلام (ASR) ،يتم دراسة الأداء على اللغة الإنجليزية أو على  اللغة العربية الفصحى. لكن في الحياة اليومية، نادراً ما يتحدّث أحد بالفصحى! هذا ماقاله الدكتور سانجيكا هيوافيثارانا، رئيس قسم حلول الذكاء الاصطناعي في aiXplain اللهجة المصرية تعتبر أكثر انتشاراً عالميّاً ، لكنها  سجّلت أعلى معدلات خطأ بين اللهجات الأربعة (الشاميّة ، الحجازية ، النجديّة ، المصريّة)  وبرغم من كثرة توافر المحتوى باللهجة وجدنا أنها تواجه عدة تحديات : في ضعف النمذجة الصوتية ضعف مع الموسيقى غياب معيّار موحد للكتابة العاميّة يكُمنُ السؤال العلمي : ليس أيّ نموذج ASR تختار. السؤال أن تُقيّم قبل أن تُطلق. تصفّح تقرير تقييم نماذج التعرّف الآلي على الكلام للّهجات العربيّة :linkedin.com/posts/aixplain…

Throwback to CNBC’s Dan Murphy interviewing our CEO Hassan Sawaf on Access Middle East, on the future of enterprise transformation and the role of AI orchestration in shaping that future. “We have today 55,000 agents built on aiXplain. We have 40,000 assets which are the building blocks of these agents, and you can build very complex systems in a simple form.” @Hassan_Sawaf , CEO, @aixplain Read full interview here: aixplain.com/blog/aixplain-…

@aixplain member Nur Hamdan, Head of Product, live from AI Agent Conference 2026 in New York. If you’re around and want to connect, reach out and we’ll put you in touch!

اكتشفنا أنّ معدل الخطأ المرتفع WER لايعبّر ذلك عن عدم نجاح النموذج ، بل يعكس اختلافات في الكتابة الغير الموحدة. ..التحسينات تشير إلى أن العديد من الأنظمة يمكنها استيعاب معنى الكلام بشكل صحيح ، لكنها تواجه تحديات في الاختلافات بين اللهجات. ~سانجيكا هيوافيثارانا ـ رئيس العلماء التطبيقيين يشمل تقريرنا مايلي: 1. قياس الأداء على اللهجات (الشامي ،المصري ،الحجازي ،النجدي) 2. تقديم مقياس لحل مشاكل توحيد الكتابة تحت اسم : editedWER 3.قياس التكلفة وسرعة الاستجابة لتطبيقات عمليّة حقيقة قم بتحميل تقرير المعايرة "التعرف التلقائي على الكلام باللغة العربية المحكيّة" : linkedin.com/feed/update/ur…

Our new benchmark found that high WER often doesn't mean failure. It means non-standard spelling: ..improvements through normalization suggest that many systems capture the meaning correctly, but have challenges with the surface-level variations. ~Sanjika Hewavitharana, Principal Applied Scientist, What we did: Benchmark on real-world dialects (Chami,Egyptian, Hijazi,Najdi ). Introduced editedWER to fix normalisation issues. Included Cost & Latency for practical use. Download the new Arabic ASR Benchmark Report: linkedin.com/posts/aixplain… #ArabicAI #ASR #aiXplain

5 papers. 3 U.S. patents. 3 months. This is what it looks like when research and product move together, with customers close to the process. Proud of the team building this 👏 Checkout aiXplain research here: aixplain.com/research/

Python code integration in aiXplain Agents allows you to add your custom code directly in aiXplain Studio, giving your agents the specific functionality they need to scale. Import your Python code today and create tailored solutions that meet your requirements. Visit Studio.aiXplain.com and give it a go!

New to MCP? Model Context Protocol (MCP) is an open standard that lets AI agents connect with external tools, APIs, and data sources seamlessly. Visit aiXplain Studio here: studio.aixplain.com/?utm_source=X&… In this video, we walk you through how to connect an MCP to @aiXplain Agents inside aiXplain Studio, step by step, so you can start building more powerful AI workflows straight away.

Announcing aiXplain Studio a no-code way to design autonomous AI agents that are governed and production-ready from day one. You define the goal. aiXplain Studio chooses and orchestrates the right tools at runtime, adapts as conditions change, and stays observable and compliant throughout. We kept seeing the same problem: teams with real problems to solve, spending their time drawing flowcharts instead of building. Wires, nodes, router logic, tools that ask you to think like a machine, not like someone who has a job to get done. So we built something different. Define the goal, not the route. Attach your tools. Set your rules. aiXplain Studio handles the rest: planning, executing, self-correcting, and keeping a full trace of everything it does so nothing runs blind. Our Head of Product Nur Hamdan wrote about what went into it and why it's built the way it is. Check Product Hunt here: #comments" target="_blank" rel="nofollow noopener">producthunt.com/products/studi… If you've been waiting for agentic AI that actually fits the way your team works, this is it. Start free today at studio.aixplain.com/?utm_source=X&…

3/3 Customers are already building production systems with aiXplain Studio, including: • Internal Research Synthesizer • Competitive Intelligence Watcher • Sales Qualifier Agent • Contract Compliance Reviewer • Financial Reporting Agent • Clinical Transcription Validator • Executive Assistant Agent Learn more about how aiXplain Studio works and why we built it this way. studio.aixplain.com/?utm_source=X&…

2/3 @aixplain Studio supports both business users and developers, with a seamless path from no-code design to deeper control when customization is required.

1/3 Introducing @aixplain Studio 🚀 a no-code platform for designing autonomous, governed AI agents that are ready for production from day one. try product here: producthunt.com/products/studi… 🚀Define goals, not rigid workflows ⚙️Select and use tools dynamically at runtime ⏲️Recover, adapt, and improve over time 🦾Stay governed, observable, and compliant in production

“AI is expensive. Subscriptions alone won’t pay for it. So the next era is going to be funded by the oldest engine on the internet: advertising.” @monamhamdy, Chief Ethics Advisor at @aixplain “Except this time, ads won’t just target demographics. They’ll target you. Not the you you post online. The “you” you confess to a machine at 1 a.m.” read full article here: