Ratul Minhaz

We are pleased to announce that LughaatNLP has surpassed 100,000 downloads across all released versions. LughaatNLP was developed with the objective of providing a robust, reliable, and standardized Urdu pre-processing library, addressing a critical gap in natural language processing resources for the Urdu language. This milestone reflects the growing adoption of LughaatNLP within the research and developer community and reinforces its role as Pakistan’s first comprehensive and dependable Urdu pre-processing framework. We extend our sincere appreciation to researchers, developers, and contributors whose feedback, usage, and support have been instrumental in this achievement. We remain committed to advancing Urdu NLP research and enabling reproducible, high-quality language technologies. #LughaatNLP #UrduNLP #NaturalLanguageProcessing #OpenSourceResearch #AIResearch #ComputationalLinguistics #PakistanTech

45

3 Comments

🚀🚀From self-learning to system-building. When I started learning PSX (Pakistan Stock Exchange) for my investments, I relied on videos, articles, and scattered websites. Over time, I consolidated that learning into an AI-powered research tool. The tool analyses 5 years of PSX data per company and evaluates stocks across multiple dimensions: 🗝️ Fundamentals (primary driver) 🗝️ Management quality & sector performance 🗝️ Macroeconomic signals 🗝️ Market sentiment & technical 🗝️ Qualitative factors (yes, even institutional stability signals) The output: a shortlist of long-term investment–worthy companies aligned with my investment mindset. It’s far from perfect. There’s significant room for enhancement. Today, it thinks like me. Tomorrow, with iteration and refinement, the goal is to evolve it into a top-tier AI investment advisor. Building. Learning. Compounding. 🚀Attached is the main workflow responsible for shortlisting companies. It’s supported by multiple helper workflows that feed it structured data. 🚀 Each deep-research execution costs PKR 500–600 (100 companies), as it pulls live data (live only for today, past data is cached), covering the period from today back to the last 5 years, to ensure decisions are based on current and historically grounded signals.

17

6 Comments

Tip: Scale large HNSW vector sets in Redis by splitting your dataset across multiple instances and querying them in parallel with multiplexing, enabling efficient parallel reads and writes that drastically improve performance and manage memory effectively. This approach leverages Redis's scalable vector support to handle hundreds of millions of vectors seamlessly. #Redis #HNSW #VectorSearch #ScalableAI #DataScience

1

BigQuery is Google Cloud’s fully managed, serverless data warehouse built for lightning-fast SQL analytics. It handles massive datasets efficiently using columnar storage and parallel processing. With built-in ML, real-time streaming, and seamless GCP integration, BigQuery powers modern data platforms at scale.

4

Your next teammate might be an AI Agent. But how do you trust it? AI is moving from a simple tool to an autonomous teammate. This raises a critical question that most aren't asking: How do we manage and secure these agents? In the first episode of our new series, "IAM for Agentic AI," our VP & GM of IAM, Geethika Cooray, breaks down this evolution and introduces a new paradigm.

12

One tool that’s been catching my attention recently in the data ecosystem is Apache DataFusion — a high-performance SQL query engine built on Apache Arrow and written in Rust. Traditional data processing frameworks often rely on JVM-based execution layers. DataFusion takes a different approach by using Arrow’s columnar in-memory format to enable extremely efficient analytical query execution. A few technical aspects that stand out: • Vectorized query execution optimized for analytical workloads • Built on Apache Arrow, enabling zero-copy data sharing across systems • Lightweight embedded query engine that can power analytics services and custom data platforms • Works seamlessly with Parquet and modern lakehouse architectures From a data engineering architecture perspective, DataFusion is interesting because it acts more like a building block than a monolithic framework. Many emerging data platforms are beginning to use it to create custom query engines, real-time analytics layers, and next-generation data services. As the ecosystem moves toward modular lakehouse architectures, tools like DataFusion highlight how high-performance data infrastructure can be built with smaller, composable components. Curious to see where Rust-powered data systems go next in the modern data stack. #data #DataEngineering #ApacheDataFusion #ApacheArrow #Lakehouse #Rust #Parquet #BigData #Analytics #CloudData #c2c #w2 #c2h @Lakshya Technologies Allegis Group Randstad USA Adecco ManpowerGroup Robert Half Kelly Science, Engineering, Technology & Telecom Aerotek TEKsystems Insightglobal Kforce Inc Spherion ExpressEmploymentProfessionals-Parma, OH Pinnacle Group, Inc. SignatureConfirm Collabera Digital Michael Page Inc. DigitalTech Media Modis Vaco Apex Systems Yoh, A Day & Zimmermann Company DISYS CyberCoders Volt Aston Carter Accountemps and OfficeTeam, Grand Rapids, MI PROSTAFF Beacon Hill Ajilon Executive Nederland ettain group Synergis Addison Group Brooksource The Judge Group Red Oak Technologies, Phaxis, Mindteck,Pyramid Consulting, Inc Amida Group

3

Dear All - Salaam Stanford University have made their entire AI curriculum freely available to everyone. Usually more than $250,000 for these courses combined. These aren’t watered-down versions. These are the exact same courses taught by Stanford professors, fully recorded and accessible to everyone. 1.⁠ ⁠CS221 - Artificial Intelligence Principles and Techniques Foundations of AI, search algorithms, reasoning, planning, and decision-making 🔗 https://lnkd.in/dA3b4ZYr 2.⁠ ⁠CS229 - Machine Learning (Andrew Ng) The legendary course on supervised/unsupervised learning and core ML algorithms 🔗 https://lnkd.in/dn6jVpEv 3.⁠ ⁠CS230 - Deep Learning (Andrew Ng) Neural networks, CNNs, RNNs, and production-level deep learning practices 🔗 https://lnkd.in/dS9htVma 4.⁠ ⁠CS229M - Machine Learning Theory The mathematical foundations and proofs that power ML 🔗 https://lnkd.in/d4csbhVQ 5.⁠ ⁠CS234 - Reinforcement Learning MDPs, Q-learning, policy gradients, and decision optimization 🔗 https://lnkd.in/d99yGGu4 6.⁠ ⁠CS224N - NLP with Deep Learning Transformers, embeddings, and state-of-the-art language architectures 🔗 https://lnkd.in/dSkR_9-n #YouTube #StanfordUniversity #Advance #NLP #DeepLearning #MachineLearning #AI #CNNs #RNNs #ProductionLevelDL #Free

249

24 Comments

Just released Kani-TTS-400M-UR, the smallest Urdu Text to Speech model built upon Kani-TTS 400M, trained on an Urdu speech dataset and optimized for low resource language conditions. Model: https://lnkd.in/eNmwbD-h Inference: https://lnkd.in/eHxBg6qv

46

6 Comments

Building in public comes with real lessons. After sharing AivioraOps, a few engineers tried the live demo and pointed out a backend issue affecting login. That feedback mattered. The issue has now been fixed, and the demo is working as expected. Big thanks to Syed Shahzaib Zafar and Mubashir Nanath for taking the time to test and share feedback — this is exactly what helps improve real products. This is what building production systems looks like: • Not just writing code • Handling real users • Fixing real issues quickly • Improving continuously Shipping is one part — responding fast is what matters. If you haven't tried it yet, you can check it here: https://ops.aiviora.online Demo credentials: Email: [email protected] Password: Retail@123 Feedback is always welcome. 🤍

23

4 Comments

“Ye AI kya insaano ko replace kar dega?” 344 days in, multiple models and agents running side by side, and one clear answer: AI doesn’t replace people. It replaces slow workflows, guesswork, and busywork. AI is there to stay. From Pakistan to global clients, the edge wasn’t AI itself, it was knowing how to think, decide, and build with it. The future isn’t humans vs AI. It’s humans with AI outperforming everyone else.

9

3 Comments

👨‍💻 Built a real-time transcription system using Whisper + LiveKit 🎙️ Added speaker detection, conversation labeling, and smooth UI ✅ Everything worked perfectly in dev… 🚨 Then came production: The user spoke Urdu over bad Wi-Fi in a noisy room 😵‍💫 And suddenly, the AI forgot what words were. Real-time STT isn’t just a model — it’s chaos control. If you’ve shipped AI to production, you know the pain. 😅 #SpeechToText #WhisperAI #LiveKit #RealTimeAI #MLEngineer #AIInProduction #LLM #MachineLearning #LinkedInHumor

3