Sushmitha Krishna Kumar

Proud to have contributed to our collaborative project with Google Cloud Security on open-source software security! We've developed key tools like PROBER and a Vulnerability Prioritisation approach to strengthen the software supply chain. #SoftwareSecurity #SupplyChainSecurity #VulnerabilityManagement #CSIRO #Google

10

1 Comment

Does your data make traditional clustering approaches like K-means less effective? Rukshan Pramoditha proposes an alternative: spectral clustering. Read his explainer to understand how (and why) it works.

12

Meta’s REFRAG: rethinking RAG decoding Just read Meta Superintelligence Labs’ new paper on REFRAG — a smart decoding framework for Retrieval-Augmented Generation that compresses retrieved passages, senses which chunks matter, and expands only the important bits during generation. The result is striking: up to 30.85× faster time-to-first-token, roughly 16× larger effective context, and lower token usage — without loss of accuracy across standard RAG benchmarks. Why this matters: real-world RAG systems (search+LLM) often jam huge retrieved contexts into the model, paying heavy latency and memory costs for mostly irrelevant tokens. REFRAG changes the execution pattern: do cheap, compressed checks for most chunks and only expand the handful that actually contribute — the win is both speed and scale. Practical implications: faster conversational agents, cost-efficient enterprise RAG at scale, and the possibility to run deep long-document reasoning with smaller LLMs. Meta will release code (and community ports already exist) — worth experimenting with for any production RAG pipeline. #RAG #LLM #AI #ML #NLP #AIEngineering #MetaResearch source : https://lnkd.in/gPuBpvES

3

Excited to announce that Studitory is partnering with CSIRO's Data61, working closely with Xiwei (Sherry) Xu and the Applied AI systems team to develop a cutting-edge AI marking workflow to mimic real teacher marking patterns! This collaboration will enable us to fully automate the exam practice experience on Studitory, making learning smarter, more reliable and more personalized than ever. A huge thank you to Sherry and the Data61 team for joining us on this journey, excited to see how we can transform the learning space! #AI #EdTech #Innovation #Studitory #Data61 #Collaboration

30

1 Comment

As AI systems continue to evolve toward autonomy and collaboration, frameworks like #CrewAI are becoming increasingly relevant. This course offered a solid overview of building and managing multi-agent workflows, with practical insights into agent orchestration, role assignment, and advanced coordination strategies. Even for those with experience in AI engineering, CrewAI presents a compelling way to experiment with agentic architectures in both prototyping and production contexts. A valuable addition to the toolkit for anyone working on intelligent, scalable AI systems. #AI #MultiAgentSystems #CrewAI #AgenticAI #ArtificialIntelligence #AIEngineering #LLM #AutonomousAgents #AIEngineer #CrewAi

27

2 Comments

I recently had the opportunity to sit with the team at Mentoria for a detailed conversation on the evolving landscape of AI and Machine Learning. The discussion examined how unexpected opportunities emerge in this rapidly developing domain and why certain underlying competencies often become decisive factors for recruiters evaluating talent. We explored foundational questions that many beginners overlook—what remains unsaid when entering the field, why successful careers extend far beyond coding alone, and which skills retain their relevance as the technological ecosystem shifts. The interview also includes a rapid-fire segment and a focused look at three specific CV issues that can lead to immediate rejection, underscoring the importance of precision and clarity when presenting one’s work. The full interview is available here: https://lnkd.in/gBV4J-TR #AI #MachineLearning #DataScience #CareerTips #Innovation #TechCareer #FutureSkills #Interview

7

Data and AI Leadership: Navigating the 2026 Tech Reset As we reach the middle of February 2026, the Australian market is undergoing a significant reality check. The narrative has shifted from the broad potential of AI to the specific, measurable disruption of existing business models and the infrastructure required to sustain them. Three developments from the last few days that caught my eye: The ASX Software Rout: We are witnessing a fundamental repricing of the Australian tech sector. Major players like Xero and WiseTech have faced significant pressure as the market grapples with the rise of Agentic Coding. Tools such as Anthropic’s Claude Code are challenging the traditional per seat SaaS pricing model. For leaders, this highlights the urgent need to move beyond simple software subscriptions and toward proprietary, value based AI integrations. The Victorian AI Mission Statement: The Victorian Government has formally launched its vision to become the nation's AI leader. This strategic framework aims to generate $30 billion in Gross State Product over the next decade. Backed by an initial $13.6 million in funding, the plan includes a $5.5 million Sustainable Data Centre Action Plan and an $8.1 million AI Career Conversion program. For Victorian leaders, this signals a shift toward state supported, responsible AI adoption. The Global Distillation Debate: In a significant move for model integrity, OpenAI has officially accused DeepSeek of "adversarial distillation". The memo to US lawmakers on 12 February 2026 alleges that the Chinese firm bypassed API restrictions to "free-ride" on frontier model capabilities. This development underscores the growing importance of Governance Frameworks to protect intellectual property as AI begins to "eat itself" through model to model training. The Bottom Line: This week’s volatility is a reminder that AI is no longer a speculative future: it is an active disruptor of legacy revenue streams. Success for organisations in Victoria and across Australia now depends on technical resilience and a commitment to workforce upskilling. #DataLeadership #AIVictoria #AgenticAI #SaaSDisruption

1

Hi All, LangGraph Workflows and Agents explores the fundamentals of agent-based systems and workflow using LangGraph. This blog covers the basics of agents and traditional workflows, and walks through key workflow patterns such as prompt chaining, routing, parallelization, orchestrator–worker, and evaluator–optimizer. With clear explanations and practical examples, the post helps you understand when to use workflows versus agents and how to design scalable, intelligent LLM applications. https://lnkd.in/gUh_vYYT Thank you, Krish Naik for motivating me to share my learnings. If you find this blog helpful, please follow me on Medium.

23

I have been working on real-world problems recently, and one thing is clear. Organisations often look for AI or IT experts to solve problems. But the best solutions come from understanding the domain first, not the technology. AI and IT are powerful tools, but they support the solution, they are not the solution. The real impact comes when domain knowledge meets the right technology. What do you think? 🤔

19

2 Comments

Recently, while scrolling through Machine Learning Street Talk episodes on Spotify, a title accusing modern AI of being an ‘imposter’ caught my eye. The episode features guests Kenneth Stanley and Akarsh Kumar discussing their recently published paper 'Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis'. At a high level the paper explores an alarming question: what if AI models like ChatGPT are imposters, faking advanced reasoning and understanding through impressive performance without actually forming a coherent internal model of the world? Stanley and Kumar argue that the AI industry is representationally optimistic: we boldly assume strong performance implies true internal understanding. The Fractured Entangled Representation Hypothesis (FERH) challenges this idea. FERH hypothesizes that deep learning models trained with backpropagation and SGD often succeed at reasoning tasks without forming elegant, modular representations of concepts. Instead, their internal representations are fractured and entangled, with the same concepts redundantly represented across many parts of the network. Instead of having a clear representation for ‘cat’, a model may rely on several loosely connected features tied to different images and contexts of ‘cat’ and often uses different pathways through features to construct the same concept. This is because SGD only rewards correct answers without taking the path to the answer into account. Stanley and Kumar equate this to two students taking a math exam. One student truly understands calculus; the other memorized the test problems. Both can ace the test but only one will likely be able to further the field of mathematics. Today’s AI systems are like the memorizer, and by expecting them to be autonomous and creative, we are asking them to develop conjectures or prove a new theorem. We can already see the cracks in the model’s façade of understanding. Ask an image model to only change your hair color in a photo, and it may very well alter the background or the color of your shirt. This is because the model struggles to isolate ‘hair color’ in its representation and tugs on other representational fragments while trying to piece together the ‘hair color’ concept.   There is no denying that AI seems promising, but FERH raises a critical concern: performance alone may be masking a tangled representational mess, devoid of the understanding necessary to create the autonomous and truly intelligent models we seek. Give it a read (or a listen) 👇 🎧 MLST episode: https://lnkd.in/edPZyrFb

20

4 Comments

Data is no longer something we store — it’s something we design. Today, I explored how the way we design and choose databases decides how our systems think, scale, and evolve. Reflections on Today’s Session- Comparative Study of Traditional & Contemporary Databases for Modern Application Development Speaking to a diverse audience of Tech & Non-Tech Background People, I experienced firsthand how meaningful learning happens when complexity meets clarity. Explaining relational vs. non-relational systems, NoSQL and Graph models, and Polygot Persistence reminded me that teaching technology is about why it matters, not just how it works. Key Insights- 1. Evolution of Databases From hierarchical and relational models to cloud-native, distributed, and AI-augmented systems, databases have evolved from passive storage to active intelligence layers powering modern applications. 2. Basis of Segregation Traditional RDBMS (Oracle, MySQL, PostgreSQL): Focus on ACID compliance and strong consistency. Modern Databases (MongoDB, Cassandra, Neo4j, CockroachDB): Emphasize horizontal scaling, flexibility, and eventual consistency, guided by the CAP Theorem trade-offs — balancing Consistency, Availability, and Partition Tolerance. 3. Major Categories Document (MongoDB): Semi-structured, schema-flexible data Graph (Neo4j): Relationship-driven analysis Time-Series (InfluxDB): Real-time IoT and telemetry Columnar / Key-Value (Cassandra, Redis): Big data and caching NewSQL (CockroachDB, TiDB): SQL reliability with NoSQL scalability Vector Databases: The foundation of GenAI and semantic search 4. Hybrid “Polygot Persistence” Modern systems blend SQL for reliability, NoSQL for speed, Graph for connectivity, and Vector for intelligence — optimizing across CAP trade-offs to meet domain-specific needs. 5. Domain-Driven Choices FinTech: SQL / NewSQL for transactional integrity Healthcare: Document + Graph for patient data IoT: Time-Series + Key-Value for streaming Gaming / Social: Graph + In-Memory for responsiveness AI / ML: Vector + Columnar for embeddings and inference 6. Benchmarking & Scalability Using TPC-C, TPC-DS, and YCSB benchmarks, we analyzed latency, throughput, and elasticity — reaffirming that cloud-native distributed databases outperform traditional monoliths at scale. 7. Future Trend Curve The next era belongs to AI-native, autonomous, vector-aware, and quantum-resilient databases — systems that can self-optimize and evolve with data. My Opinion- True database mastery lies in design thinking, not syntax. When learners begin to ask why a data model exists, they start thinking like architects. Today’s discussions reaffirmed that the next generation must be data-strategic, not just data-driven — viewing databases as the core of digital intelligence, not merely storage.

43

Having trouble with Retrieval systems ("R" in RAG Systems), mining GBs of text for the correct information chunk for output generation? You are not new! Given millions of chunks, finding the top 5 is a challenging problem (with ~99.995% ranking accuracy). Would an LLM check all the chunks and find the best ones, spending days doing so? The solution lies in building an efficient "Engineering solution" driven by an Agentic AI Multi-hierarchical Retrieval System !! Typical Solution would be -- Millions of chunks --> Organized by key information (Hierarchical Knowledge Graphs, etc.) --> Identifying key top 100-1000 relevant chunks --> Using Agentic AI to generate outcome. Benefits of this approach-- 1. Bring down the accuracy requirements for retrieval systems by better organizing data 2. Leverage LLMs to share some grunt work Costs of this approach-- 1. ~5-30 mins of processing time 2. ~$10-$100 of the cost of generation Is it worth it? Have you experienced this problem yet?

21

5 Comments

This week Hema Manasi Potnuru and I explored the roles of MCP, LangChain, and LlamaIndex—three names that often come up together but operate at different layers of the AI stack. In the book Fundamentals of Data Engineering, Joe Reis and Matthew Housley Housley emphasize the importance of understanding each layer of a data or AI system before choosing tools. With that perspective, we looked at these emerging and promising technologies—LlamaIndex, LangChain, and MCP—and explained where each fits within the layers of a modern data or AI system. MCP – the protocol layer standardizing how LLMs connect with tools and systems. LangChain – the orchestration framework for workflows, agents, and memory. LlamaIndex – the data integration layer for connecting LLMs with enterprise data sources. We also outlined their overlaps, limitations, and when to use each: Quick prototypes → LangChain Complex enterprise data → LlamaIndex Future-proof infrastructure → MCP Our key takeaway: framework fatigue is growing, and while each has its place today, MCP may become the foundation that unifies them. Read the full breakdown here: https://lnkd.in/gBHDAmHu #AI #MachineLearning #DataScience #DataEngineering #LLM #LargeLanguageModels #KnowledgeManagement #VectorDatabases #LangChain #HuggingFace #OpenAI #AmazonWebServices #GoogleCloud #TechInnovation #GenerativeAI #AIApplications #IU #IUB #Data #ChatGPT

14

🇦🇺 Australia’s AI Governance at a Critical Juncture This week marked a pivotal moment for AI governance as the UN established new coordination infrastructure through scientific panels and global dialogue mechanisms – setting the stage for how nations like Australia will navigate AI oversight. 1️⃣ The Challenge: Research reveals widespread “shadow AI” adoption across Australian workplaces, where employees use unauthorized AI tools, creating vulnerabilities in data security and compliance. This disconnect between institutional policies and actual workforce behavior highlights a fundamental governance gap. 2️⃣ The Opportunity: Australia’s government has successfully implemented AI assurance protocols based on national AI Ethics Principles, proving that principled governance can coexist with innovation. Meanwhile, recent policy discussions show potential for breakthrough frameworks that balance union consultation requirements with business productivity needs. 3️⃣ The Insight: Effective AI governance isn’t about constraining innovation through rigid regulation – it’s about developing adaptive frameworks that evolve with technology while maintaining essential safeguards.Organizations that proactively address shadow AI risks and develop governance capabilities will lead in the AI-integrated economy. Those that delay face mounting compliance and competitive challenges. The path forward requires understanding that AI governance operates across interconnected local, national, and international systems. How is your organization preparing for this new governance landscape? #AIGovernance #Artificalintelligence #AI #Datascience #businessintelligence #digitaltransformation

6

2 Comments

AI is the buzzword of the decade, but is it truly transforming the world, or just riding the hype wave? In this ML Bites post, Kaveesha Vidushinie dives into the heart of the debate: exploring what makes AI revolutionary, where it stands today, and whether it’s living up to the promise. https://lnkd.in/gtTJt2sD

6