Mahnaz Asghari

🚀 Excited to share my project: Smart Persian Product Search! Finding the right product on Persian e-commerce platforms is often challenging due to: 🔸 Different naming conventions 🔸 Inconsistent attributes like color, size, and material 🔸 Traditional search engines failing to combine semantics with structured attributes To tackle this, I built a semantic & attribute-based search engine that integrates: ✨ Semantic Search – Sentence Transformers for query understanding ✨ Attribute Matching – gender, size, color, material, subcategory ✨ Keyword Boosting & Scoring – TF-IDF for term relevance ✨ Synonym Mapping – handling Persian variations ⚡ FAISS Indexing – for fast similarity search 🛠 FastAPI REST API – seamless integration with e-commerce platforms This system helps Persian online shops deliver faster, more accurate product discovery, even with incomplete or ambiguous queries. 🔗 Check out the full project on my GitHub

27

🚀 Thrilled to Announce My Latest Research Publication! 📢 I'm excited to share that my research paper, "Hadoop to BigQuery: Migrating Automotive Data Lakes Without Downtime", has been officially published in The American Journal of Interdisciplinary Innovations and Research (Vol. 07, Issue 07). - Read it here: https://lnkd.in/ezJHtmTm - DOI: 10.37547/tajiir/Volume07Issue07-03 As the automotive industry generates massive, fast-evolving datasets—from connected vehicles to manufacturing sensors—the need for scalable, real-time data infrastructure is more urgent than ever. This paper dives into how organizations can modernize their data lakes with Google BigQuery while ensuring zero operational downtime. 🔍 Top Highlights: - Zero-Downtime Migration Techniques: A practical roadmap using blue-green and canary deployments, real-time data replication, and hybrid strategies tailored for mission-critical automotive systems. - Hadoop vs. BigQuery – A Modernization Perspective: An in-depth comparison showcasing how BigQuery's serverless architecture, streaming capabilities, and built-in ML tools outperform traditional Hadoop setups for automotive use cases. - Real-World Case Studies: Learnings from industry giants like Uber and Geotab on scalable cloud migrations and performance gains using Google Cloud’s ecosystem. I hope this research adds value to professionals working on data modernization, analytics, and AI/ML infrastructure—especially in high-volume sectors like automotive and manufacturing. #DataEngineering #BigData #GoogleCloud #BigQuery #AutomotiveAnalytics #CloudMigration #ResearchPublication #MachineLearning #DataLakes #Hadoop #TechInnovation #ZeroDowntime #CloudComputing

7

6 Comments

𝐁𝐫𝐢𝐧𝐠𝐢𝐧𝐠 𝐓𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐜𝐲 𝐭𝐨 𝐀𝐈 𝐢𝐧 𝐅𝐥𝐞𝐞𝐭 𝐌𝐚𝐧𝐚𝐠𝐞𝐦𝐞𝐧𝐭 🚛🤖 Thrilled to share that my recent research paper is now published! 📢 Machine learning powers everything from predictive maintenance to driver risk analysis, but its “black box” nature limits trust and compliance. Our latest work integrates Explainable AI (XAI) tools like SHAP and LIME into the fleet ML pipeline to make models both accurate and transparent. ✅ Results: - 55% reduction in unplanned maintenance costs - Clear, human-understandable insights for regulators and operators AI shouldn’t just predict, it should explain. 🔗 Read it here: https://lnkd.in/eFR3ADbT #AI #MachineLearning #ExplainableAI #FleetManagement #PredictiveMaintenance #DataScience #XAI #Transparency #Innovation #TrustInAI

5

Custom AI training programs aren't just about delivering content, they're about transforming how your team works with AI every day. At Data Society, we design training that meets your organization where it is. Real instructors guide your team through hands-on learning tailored to your specific challenges. This is skill-building that sticks. We combine live teaching with ongoing support like our Virtual Teaching Assistant that provides real-time feedback. This ensures learning translates into measurable impact quickly. For example, a government agency we partnered with saw accelerated AI adoption by adding custom training with governance frameworks and real workflows. Their teams gained confidence and ownership over AI tools, driving better outcomes. The takeaway? Custom AI training isn't a one-size-fits-all checkbox. It's a strategic investment in your people and processes that powers responsible AI adoption at scale. How are you designing your AI learning path to create real impact?

3

I’m excited to share that our paper, “Domain-Specific SQL Generation with LLMs: A Hybrid Framework Combining Knowledge Graphs and Retrieval-Augmented Generation,” has been published in Advanced Engineering Informatics. In this work, we propose a hybrid framework that integrates knowledge graphs with retrieval-augmented generation (RAG) to improve domain-specific SQL generation using large language models. The approach enhances schema awareness, query accuracy, and robustness in specialized database environments. Grateful to my adviser Dr YONGWEI SHAN and collaborators Dr Weihua Sheng, Dr Hossein Khaleghian, and Dharmendra Reddy Chitte for their guidance, insights, and collaboration throughout this project. Free access to the paper for the next 50 days: https://lnkd.in/duzNhp7N

53

4 Comments

https://lnkd.in/gtiAqrMK What if you could perform advanced dataflow analysis without ever compiling code? Welcome to the future of program analysis. In this episode, we unpack LLMDFA, a groundbreaking framework that uses Large Language Models to perform customizable, compilation-free dataflow analysis — even on code that won’t compile. 🔍 We discuss: The limitations of traditional dataflow analysis and how LLMDFA overcomes them How LLMs are used as reasoning engines to outsource tasks to expert tools like theorem provers and parsing libraries Chain-of-thought prompting and decomposition techniques to reduce hallucinations Real-world applications in Android bug detection and synthetic program testing Why LLMDFA sets a new benchmark with 87.10% precision and an F1 score improvement up to 0.35 Whether you're working on static analysis, LLM tooling, or software verification — this episode connects LLM innovation with core programming languages

28

1 Comment

An excellent paper by Anthropic on what is exactly an agent, what are the typical workflows and agents and when to use each one - https://lnkd.in/gM3t8xhp Here is an excerpt from the paper: 1.) Workflows are systems where LLMs and tools are orchestrated through pre-defined code paths. 2.) Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.

25

🚀 Understanding the Types of Retrieval-Augmented Generation (RAG) Retrieval-Augmented Generation (RAG) is transforming how we build intelligent systems by combining the power of retrieval with generative models. But did you know there are different types of RAG architectures, each suited for specific use cases? 🔍 Here's a quick breakdown: Classic RAG Retrieves documents and feeds them into a generative model. Great for open-domain QA and summarization. RAG-Sequence vs RAG-Token RAG-Sequence: Generates output based on the most relevant document. RAG-Token: Generates each token by attending to multiple documents. Multi-hop RAG Retrieves in stages to answer complex queries requiring reasoning across multiple documents. Conversational RAG Maintains context across turns in a dialogue. Ideal for chatbots and virtual assistants. Hybrid RAG Combines dense and sparse retrieval for better coverage and relevance. 💡 Choosing the right RAG type depends on your task, data, and performance needs. #AI #LLM #RAG #RetrievalAugmentedGeneration #MachineLearning #GenerativeAI #NLP

13

MIT recently published a study on cognitive debt when using AI assistants. It confirms what many suspected: brain activity is lower when you rely on an LLM to think for you. This raises important concerns. Could over-reliance on AI gradually erode our critical thinking skills? What are the implications for education and child development, where cognitive growth is essential? Interestingly, the study found that outcomes improve when you engage your brain first, then use an LLM to augment your original thinking. In other words: think first, then prompt. Link below, in case you want your LLM to summarise it for you. https://lnkd.in/dc4PSmJ7

31

1 Comment

KAPPA: parallel compute for unstructured metadata extraction for AI and other use cases! If you have petabytes of data needing industry or organizational-specific data classification and labeling, this is something to check out.

14