Ravi Ramakrishnan (.

Question : Responsible AI deployment - what is that and why should an AI product manager bother about it❓ Ethical development of AI products is not just about algorithmic transparency or handling sensitive user data. One critical aspect of responsible AI deployment is the need for ongoing monitoring and evaluation. 👉 Product managers should establish robust mechanisms to continuously assess the performance and impact of their AI models, ensuring they remain aligned with intended use cases and do not exhibit unintended or harmful behaviors over time. 👉 Product managers must consider the potential for AI systems to amplify existing societal biases or create new forms of discrimination. Proactive measures, such as conducting bias assessments and implementing fairness-aware algorithms, are essential to mitigate these risks and promote inclusive and equitable AI solutions. 👉 Transparency and accountability are also crucial pillars of responsible AI deployment. Product managers should strive to provide clear and accessible information about their AI models, including their capabilities, limitations, and decision-making processes. 💡 Quick Tip: Track your model performance not only from technical point of view but also from ethical, social and regulatory implications of your AI product. #ArtificialIntelligence #productmanagement #productmanager #DataScience 

٧

Adverse Media Screening in the UAE: Key Insights In today’s global banking landscape, staying compliant with Anti-Money Laundering regulations is paramount, especially in regions like the United Arab Emirates, where rapid economic growth has made it a hub for international trade and finance. Adverse Media Screening (AMS) has become increasingly critical for banks operating in the UAE, given the country’s commitment to combating financial crime and maintaining its reputation as a secure financial hub. What are some of the challenges of Adverse Media Screening in the UAE? 1. Volume of Data - with the sheer volume of data and news sources, screening for adverse media can be overwhelming. 2. Manual screening - it is an inefficient approach in a high-stakes environment. 3. False Positives - leads to unnecessary delays and increased operational costs as compliance teams have to sift through vast amounts of data to verify findings. 4. Language and Cultural Barriers - a multicultural population and media landscape results in adverse media being published in various languages and reflecting different cultural contexts. Suggestions for Effective Adverse Media Screening in the UAE: 1. Leverage AI-Driven Solutions: Implementing AI-powered screening tools can significantly reduce manual efforts, improve accuracy, and minimise false positives. These solutions can process large datasets from diverse sources, ensuring a more comprehensive analysis of adverse media. 2. Ongoing Monitoring: It’s not enough to screen clients during the onboarding process, it is crucial to stay updated on any adverse media reports that could emerge after the client relationship has begun. 3. Customising Risk Parameters: Banks should tailor their AMS systems to align with their risk appetite and regulatory requirements. Different institutions may face varying levels of risk depending on their client base, so the AMS system should be flexible enough to adjust to these needs. 4. Training and Awareness: Proper training ensures that the teams can act quickly and accurately when potential risks are identified. 5. Collaboration with Regulators: Staying in close communication with regulatory authorities helps banks stay ahead of emerging threats and compliant with the latest regulations. In an increasingly complex global financial environment, AMS is not just a regulatory necessity for banks in the UAE, but a crucial element of robust risk management and long term success. As the UAE strengthens its position as a global financial hub, adopting AI and continuous monitoring solutions will be key to safeguarding their reputations, and maintaining trust with both regulators and clients. Banks that embrace these innovations will set a higher standard for security and ethical banking practices. #AdverseMedia #AML #UAE #FinancialCrime #Compliance

٢٢

Model calibration is the technique of making the model probabilities representative of the true likelihood of the event without just focusing on accuracy. This can be useful in cases where actual risk scores are needed in use cases like fraud detection or chances of defaulting on the loan etc. The technique is more helpful with sophisticated models as they have more chances of probability coming out to be non-calibarted. The technique also comes in handy when multiple scores have to be combined. But how to achieve model calibration? And how to check even if the probabilities initially are calibrated or not in the first place? Here's a short video for Data Science enthusiasts I have created on the topic covering how to check if your model is calibrated or not and what techniques one can utilize or use to have a layer of probability calibration agnostic of the underlying model/algorithm. Like and subscribe for more such interesting concepts. Also, like and share over here for maximum reach. : ) Video Link: youtu.be/rG2EfFOXyg0 YT channel Link:  youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

٤

Question : Why does quoting just the accuracy numbers do not make sense for an AI model❓ Its a common practice among data science practitioners to quote accuracy of an #AI model. While it is easier to understand accuracy numbers, this practice can be misleading and fail to capture the nuances of real-world scenarios, leading to misaligned expectations among business stakeholders. 👉 Accuracy alone does not provide a comprehensive picture of a model's capabilities. It fails to account for crucial factors such as the size and distribution of the training, testing, and validation datasets, the presence of class imbalances, and potential biases. Relying solely on accuracy can mask underlying issues and lead to false confidence in the model's performance. 👉 To truly understand a model's effectiveness, it is essential to consider a broader range of metrics. Precision, recall and F1 score offer valuable insights into a model's ability to correctly identify positive and negative instances, handle imbalanced data, and discriminate between classes. 👉 However, even these technical metrics may not resonate with business stakeholders who are primarily concerned with the practical implications of the model's performance. It is crucial to translate these metrics into business-relevant terms that directly align with the product's objectives and success criteria. For example, in a fraud detection system, precision could be framed as the percentage of flagged transactions that are indeed fraudulent, while recall could represent the proportion of actual fraudulent transactions correctly identified. By presenting these metrics in a business context, stakeholders can better understand the trade-offs and make informed decisions about the model's deployment and ongoing monitoring. 💡 Quick Tip: Dont quote accuracy numbers without mentioning the size of training and testing dataset. Consider a diverse set of metrics tailored to your specific use case and business objectives. Communicate these metrics in a language that resonates with stakeholders, fostering a shared understanding and aligning expectations. #artificialintelligence #productmanagement #datascience #productmanager

٦

Question : What skillset does an #AI product manager need on his team for building an AI product ❓ Success of any AI product lies depends largely on the team involved in building it. Crafting a successful AI-powered product requires a diverse and highly skilled team. 👉 ML Expertise : First and foremost, a strong data science andML expertise is a must-have. The team should include experienced data scientists who can not only build and train robust AI models but also analyze the data, identify patterns, and extract meaningful insights. These data scientists should possess a deep understanding of statistical analysis, algorithm development, and model optimization. 👉 MLOps engineering : MLOps engineers establish the infrastructure and processes to manage the AI model lifecycle. They package, deploy, monitor, and retrain models, addressing challenges like model drift and data changes. This ensures the AI-powered features deliver consistent, reliable performance. 👉 Seasoned Software Engineering : Alongside the data-centric roles, AI product team should also possess software engineering capabilities. Seasoned software architects and developers are needed to design and implement the underlying infrastructure, ensuring seamless integration of the AI components with the overall product ecosystem. Their expertise in system design, scalability, and security will be instrumental in creating a robust and reliable AI-powered solution. 👉 Cloud Engineering and Architecture : Cloud engineers and architects design and deploy the cloud-based resources - storage, compute, serverless functions - that power the AI engine, ensuring scalability and reliability. They implement best practices for cloud security, data governance, and cost optimization. 👉 User Experience Design : The team should also include user experience (UX) designers who can translate the AI-powered features into intuitive and engaging user interfaces. These designers should have an understanding of human-AI interaction, crafting experiences that leverage the strengths of AI while addressing the needs and expectations of the end-users. They should be skilled in user research, interaction design, and visual design. 👉 Domain-Specific Expertise: To ensure the AI product addresses real-world challenges, the team should also include subject matter experts who can provide domain-specific knowledge and insights. These experts can help validate the relevance and applicability of the AI-powered solutions. By assembling a team with this diverse set of skills, AI product managers can increase the chances of creating a successful AI-powered product that not only demonstrates the potential of AI but also delivers a compelling and impactful solution to the market. 💡 Quick Tip: Regularly assess the skills and expertise of your AI product team, and be ready to adapt and expand the team as the project evolves and new challenges arise. #artificialintelligence #productmanagement

٧

Question : Why does an AI product manager need to monitor #AI model❓ Once AI model is developed and deployed, the job doesnt end there. Model encodes pattern of the data it has seen during training phase. There may be several patterns or edge cases which are not part of training data. If model encounters any such patterns in testing or production phase, it will produce results which wont make sense without giving any warning about it. 👉 Data drift is a common challenge faced by AI-powered products. Over time, the distribution of input data can shift, leading to a mismatch between the model's training data and the real-world data it encounters. This can result in degraded model performance, unexpected outputs, and even complete model failure. 👉 The consequences of data drift can be severe, ranging from suboptimal user experiences to regulatory compliance issues. Product managers must be vigilant in detecting and addressing these challenges to ensure the continued success of their AI-powered offerings. 👉 Sudden changes in metrics can indicate input data drift. Additionally, monitoring the distribution of model outputs can reveal output drift, which may signal a need to retrain or fine-tune the model. Detecting data drift early is crucial. Product managers can leverage techniques such as A/B testing, anomaly detection, and continuous monitoring to identify and address these issues. Once detected, the product team can take appropriate actions, such as retraining the model, adjusting the data pipeline, or even considering a model architecture change. 💡 Quick Tip: Integrate model monitoring into your product roadmap and allocate dedicated resources to ensure the long-term success of your AI-powered offerings. Establish a comprehensive model monitoring dashboard to gain real-time visibility into the health and performance of your AI models. #artificialintelligence #productmanagement #datascience #productmanager

٥

For Optimization problem, getting a well behaved Cost-function, is like finding a right soul-mate. Smooth, Continuous, Differentiable, Bounded, Convex.. Cost-function ! The one, which is loyal to a single Global-Minima, and never gets trapped into shady local-minimas.. Sarv-gun sampan! Just thinking about it can dilate Mathematicians' pupils, secret oxytocin and increases dopamine-levels. #optimization #statistics #datascience #funatwork

٧٢

١ تعليق واحد

How does a bad implementation of a chatbot look like! Meet uno from OneCard ! ❌It can not decipher any transaction details. ❌It can not understand any word other than the ones pre written. ❌It can not solve the issue it has not been explicitly trained on. ( I don’t know if I am overestimating their capabilities of putting a LLM based chatbot) ❌lastly, it can not transfer to a human operator even if it fails to do any resolution. By all probability, it looks like a college project or a toy chatbot shown on YouTube videos. The issue is that it is deployed as customer facing bot within the financial industry! It is not handling food delivery issues like Zomato Zia. I hope that Reserve Bank of India (RBI) would take a note for such irresponsible approach customer service when the payments scams are on the rise in the country and regulate these companies’ as they will only make the overall financial space less accessible in the name of technology advancement.

١٦

١٠ تعليق

Establishing a full DQ Framework needs to define objectives, select tools, establish governance, implement assessment processes, standardize data, ensure security, provide training, and continuously improve. Below are really important highlights on the the first and fifth steps.

٣

"Deep learning : You throw a mountain of data in one side, something mysterious brews in the middle – layers and neurons doing their cryptic dance – and voila, outcomes pop out the other side. And that middle part? That’s the classic ‘WTF’ zone, where all the magic (or occasional mayhem) happens. But hey, it works! So we just nod along like we fully understand what's going on. #Progress, they call it. #DeepLearning #AIWonders Solution? Explainable AI (XAI) – the secret decoder for the ‘WTF’ zone. With tools like LIME and SHAP, we get a peek behind the curtain, seeing exactly how each feature pulls its weight in the prediction. Imagine it as AI’s own backstage pass, where each layer finally spills its secrets and explains what it’s been up to. And for those not-so-light cases where we need deep learning’s power but still want transparency? Interpretable neural networks are making waves, taking the ‘Why This Feature?’ vibe mainstream. So maybe it’s not all magic. With XAI, the ‘WTF’ is becoming a bit more of an ‘Oh, that’s why.’ #ExplainableAI #DataScienceRevealed"

٢٦

@For researchers and PhD/DBA students 💡: Common Fit Indices for Model Evaluation 1. Chi-Square Test 📊 Purpose: Compares observed and model-implied covariance matrices. Key Point: Sensitive to sample size; small deviations in data can lead to significant results in large samples. 2. Root Mean Square Error of Approximation (RMSEA) 🛠️ Purpose: Evaluates how well the model approximates the data. Guideline: Values < 0.06 indicate a good fit. 3. Standardized Root Mean Square Residual (SRMR) 🔧 Purpose: Measures the average discrepancy between observed and predicted correlations. Guideline: Values < 0.08 indicate good fit. 4. Comparative Fit Index (CFI) ✨ Purpose: Compares the model to a null model. Guideline: Values > 0.95 indicate good fit. 5. Tucker-Lewis Index (TLI) 🌌 Purpose: Similar to CFI but penalizes for complexity. Guideline: Values > 0.95 indicate good fit. ✍️ Summarised by Dr Hatim Dawood Al-lawati (PMP®, DBA-FinTech) 📚 Reference: Schumacker, R. E., & Lomax, R. G. (2010). A beginner's guide to structural equation modeling (3rd ed.). Routledge. ISBN: 978-1-84169-890-8 (Hardback), 978-1-84169-891-5 (Paperback). #PLSSEM #StructuralEquation #modelfit #fitindices #modelevaluation ResearchGate

٣

"Excellent breakdown of text chunking techniques! 🚀 Chunking is essential for improving NLP tasks by structuring text into digestible units. Each method, from tokenization and sentence segmentation to more advanced forms like named entity and semantic chunking, enhances the precision and efficiency of data processing. For instance, named entity chunking can significantly boost information retrieval by identifying key entities, while semantic chunking helps in understanding contextual meaning. Leveraging hybrid chunking approaches can further refine data handling and boost the performance of language models. Great insights! 🌟 #TextChunking #NLP #NaturalLanguageProcessing #AI #MachineLearning #DataScience #InformationRetrieval #TextProcessing #TechInnovation #ArtificialIntelligence #TechTips #DataAnalysis #DeepLearning #TechTrends #Chatbots #SearchEngineOptimization #DocumentManagement #DigitalTransformation #DataEfficiency #ScalableSolutions #RAG"

١

How is this quick, easy and intuitive missing value imputation technique? You can impute the missing values using the idea of the conditional distribution and marginal distribution. For e.g. Let's say you are trying to impute the missing values in education. Take the relatable features like age, income etc. Now you can impute the missing education with the sub-category within education. How? 1) Min conditional distribution distance There will be vectors for conditional distribution of each sub-category of education : P(education, age, income | education = BTech) and also for missing values : P(education, age, income | education = not known). We will choose the one with the minimum distance between these two distribution vectors. let's call this d(CD) 2) Min marginal distribution distance We also should impute keeping in mind that the overall distribution of the feature is maintained. This can be obtained by minimizing the distance between two marginal distributions, one distribution without imputed values and other with imputed values included let's call this d(MD) our multi-objective function looks like this: alpha*d(MD) + (1-alpha)*d(CD) Happy Imputing !! 😃

١

𝐓𝐡𝐞 𝐃𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐜𝐞 𝐁𝐞𝐭𝐰𝐞𝐞𝐧 𝐅𝐢𝐧𝐞-𝐓𝐮𝐧𝐢𝐧𝐠 𝐚𝐧𝐝 𝐏𝐫𝐞-𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 The processes of fine-tuning an AI model and continuing its pre-training represent two distinct methods of model adaptation and improvement. 𝐓𝐡𝐞𝐲 𝐦𝐚𝐲 𝐬𝐨𝐮𝐧𝐝 𝐬𝐢𝐦𝐢𝐥𝐚𝐫, 𝐛𝐮𝐭 𝐭𝐡𝐞𝐲 𝐚𝐫𝐞 𝐭𝐰𝐨 𝐝𝐢𝐬𝐭𝐢𝐧𝐜𝐭 𝐩𝐫𝐨𝐜𝐞𝐬𝐬𝐞𝐬, 𝐨𝐟𝐭𝐞𝐧 𝐦𝐢𝐬𝐭𝐚𝐤𝐞𝐧𝐥𝐲 𝐫𝐞𝐟𝐞𝐫𝐫𝐞𝐝 𝐭𝐨 𝐚𝐬 𝐣𝐮𝐬𝐭 “𝐟𝐢𝐧𝐞-𝐭𝐮𝐧𝐢𝐧𝐠 𝐅𝐢𝐧𝐞-𝐭𝐮𝐧𝐢𝐧𝐠 - Fine-tuning employs labeled data to fine-tune the model’s parameters, tailoring it to the specific nuances of a task. This specialization significantly enhances the model’s effectiveness in that particular task compared to a general-purpose pre-trained model. 𝐏𝐫𝐞-𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 - Pre-training usually would mean take the original model, initialise the weights randomly, and train the model from absolute scratch on some large corpora. Further/Continuous pre-training means take some already pre-trained model, and basically apply transfer learning — use the already saved weights from the trained model (checkpoint) and train it on some new domain #llm #genai #generativeai #sql #python #ai #datasc‍‍‍‍‍‍‍ience #machin‍elean‍‍‍‍ing ‍‍‍‍‍‍‍‍‍#jobhelp

٢

This is a game-changer for optimizing RAG setups! 🔧 With so many variables to consider, manually finding the best configuration can indeed be overwhelming. The introduction of RAGBuilder and its hyperparameter tuning capabilities simplifies the process significantly, allowing for efficient discovery of the optimal setup. Excited to check it out and see how it integrates with SingleStore for a streamlined RAG setup. Thanks for sharing this valuable tool! #RAG #MachineLearning #DataScience #AI #HyperparameterTuning #OpenSource #TechInnovation #SingleStore

١

National Bank of Fujairah has chosen Intellect Global Transaction Banking | iGTB's eMACH.ai to to enhance its digital capabilities 📲 🏦 🏢 NBF currently serves corporates and SMEs through its network in the UAE, and provides financial services to optimise commercial opportunities AND achieve sustainable growth for their clients. 🇦🇪 ☁ This partnership ALSO marks iGTB's FIRST cloud-managed service in this region and NBF can offer their clients a COMPREHENSIVE suite of solutions 👨‍💻 Manish Maakan, CEO of iGTB commented, “By selecting Intellect Global Transaction Banking, NBF has chosen a strategic technology partner that offers scalability, operational efficiency, and the ability to grow alongside their business. NBF’s commitment to advancing its transaction banking infrastructure reflects the critical role of technology to drive growth and innovation in Commercial Banking.” He added, “This is our first deal to be hosted on eMACH.ai Cloud managed services in the region and is a significant step in UAE’s digital transaction banking landscape. Today, over 60% of the world’s best transaction banks #winwithiGTB and this deal signifies a pivotal step in cementing iGTB’s leadership position within the UAE & GCC market, underscoring our dedication to offering leading-edge digital banking solutions.” Read more on the link below 👇 Uppili Srinivasan Prashant Lalchandani Sachin Salian Shishir J. Amit Gupta Razia Ahmed Rajesh Saxena Nachu Nagappan #UAE #igtb #transactionbanking #digitalbanking #commercialbanking #GCC #cloudbanking #fintech #managedcloud

٢