Praveen Kumar G

Kalki 2898: Five lessons in data storytelling As Kalki 2898 AD gets the cash registers ringing for Indian cinema,I was wondering if there are lessons for data professionals in storytelling from this magnum opus experiment. 1.Kalki is the 10th reincarnation of Vishnu- Kalki universe waits to unfold like the Marvel and DC universe and the world really deserves to know many captivating tales from the Indian folklores. As with the Dasavatars(reincarnations),in every well delivered data project, data blueprint comprises ten components which are the seeds of all potential revenue streams. Dasavatars of Data: Strategy,Governance, AI,Analytics,Architecture,Data products,Integration,Tools,Visualization & Downstream. 2. Remixes and retold stories told well create Gods- Elvis, Beatles, Christopher Nolan,RD Burman, Andy Warhol, Quintine Tarantino -all told someone else’s stories partly/ fully and told them with elan. Kalki is an honest (albeit still a long way to catch up)pastiche of greatness -there is no other movie in history of cinema that creates a heady cocktail of science fiction - a rum punch of Metropolis (Fritz Lang’s 100 year old magnum opus that inspired Bladerunner sets), Batman/ Batmobile,Mad Max, Bladerunner(2049- the child bearing robots/ Decard and robot Rachael’s child? Tyrell corporation headquarters), Firefly/Serenity (the greatest scifi bounty hunter to match Han Solo)? Total Recall,Dune,Matrix, Star Wars ,Terminator(remember John Conner-the future leader of the resistence, now a child, whom Arnie comes down to kill (as predicted by AI) & later protect. I always wondered was the “Terminator”story inspired by Krishna’s birth story (Krishna was predicted to kill Kansh by an Akashvani(Celestial voice)so Kanska wanted his birth stopped & now Kalki comes a full circle with a similar story? Likewise, for the data aficionado- Is it Warehouse/ DataLake or Lake house?Cocktail rules here too- Datalakehouse. 3. The rock star rules - Amidst all the razmatazz, paparazzi and glitterati going gung ho or huffing and puffing at the poor middle part of the film - Aswathama (Amitabh) stands tall - taller than Big B himself! Likewise, Without solving the buying personas pain there is no analytics -Aswathama is obsessed with one objective. 4. True to Bollywood, Kalki serves different audience classes wrt action, romance, emotion & drama: the very construct of data fabric/data products. 5. The foresight that is waiting to unfold in part 2-Is Supreme Yashkin(Kamal Hassan) actually Kali (watch how his hands unfold in the end) or Is he a repackaged Dr Octopus (Dr Otto Octavius - Doc Ock)with eight hands ? Is Vishnu the father of Sumathi (SUM80/ Deepika’s) child Kalki - is he the Krishna whose face is never shown- Krishna afterall is an avatar of Vishnu ? A great data story reveals a lot, but conceals more to keep the audience (C suite / end customer ) nail biting at the edge of a cliff hanger who always Dil Mange More … more to follow in the sequel

29

5 Comments

Wondering if it's the right time to invest in plots in Hyderabad? Here’s your answer. Learn more about investing in plots in Hyderabad by watching the full episode on plots in our YouTube channel, link to full episode in comments.  Featuring Pradeep Maganti who has over a decade of experience in the plotting space. #hyderabadplots #plotsinhyderabad #hyderabadrealestate #plotsforsale

6

1 Comment

Chronology exaggeration is mostly psychology than data driven. Aryan Invasion Theory (AIT) compressed Indian Chronology to absurd levels with Rgvedic composition period dated to 1500-1000 BCE, other Vedic texts upto 800 BCE, Mahabharata to 600 BCE and Ramayana to 400 BCE. When a ball is forcefully pushed down, it forcefully bounces back! Thus, some started proposing Rgveda in 24,000 BCE, Ramayana in 12209 BCE and Kurukṣetra War in 5561 BCE etc. The height of the bounced ball went through the roof! They were able to date like this, because no cross comparison of the information in Rgveda or Ramayana or Mahabharata was done! Had they done it, they would have found the numerous connecting links that connects all these texts like pearls in a chain. It won't allow date exaggeration. Santanu prensent in Rgveda 10th Mandala of Rgveda means there are only few generations between the end of Rgveda and the Kurukṣetra War (Santanu - Vichitravirya- Pandu - Pandavas - Abhimanyu and Parikshit).Here Parikshit was born one year after Kurukṣetra War. Rama present in 10th Mandala of Rgveda means there are no large number of generations between Rama and Kurukṣetra War (Rama - Kusa - Kusa's son - Brihadbala. Here Brihadbala was killed in the Kurukṣetra War. This is why there are large number of individuals like Parasurama, Hanumat, Markandeya, Durvasa, Vibhishana, Vaisravana, Jambavat, Mainda, Dvividha and others who lived during the period of both Rama and Pandavas. King Kuru the ancestor of Santanu is the one who invented the names Kurukshetra and Kurujangala for the regions north and south of Sarasvati ti the west of Yamun. Kurujangala is mentioned in Ramayana as contemporary place during Rama's lifetime. This also mean Rama lived later to King Kuru, similar to Santanu. This is why both Rama and Samtanu are mentioned as contemporary to 10th Mandala in Rgveda. The verb kuru (to do) itself is Late Rgvedic verb replacing the older verb krnu (to do). But large masses of people are extremely fond of chronology exaggeration. It is like adding sweet into a dish. Sweetness is a pleasant positive taste and 99% people don't mind extra sweet in their dishis. Though they are repulsed by the bitterness of AIT's chronology contraction, they don't mind the sweetness of chronology exaggeration , even though both paradigms destroy consensus and arriving at the true chronology! It is like giving an overdose of sugar to diabetic patients to make them more sick. Those who want to see our chronology remain unsettled for ever, can even think of funding for the propagation of exaggerated chronologies.

5

2 Comments

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

7

The Govt asks QCommerce firms to prove their claim of delivering within 10 minutes. Imagine you are a Data Analyst working as part of Blinkit's data team. How would you approach this problem? What questions would you ask your data? What is the right metric to work with? What concepts would be required to attempt this problem? Welcome to the world of Descriptive and Inferential Statistics with Hypothesis testing being the key to the lock. Remember? When p is low, null should go! Discover another such scenarios across industry domains like E-commerce and Insurance : https://lnkd.in/dkjASWqS Whether you are in individual data professional or are leading high performing data teams, getting good grasp on the essentials of Statistical Thinking is critical to enable them solve real world problems like this. Discover our 4D framework of upskilling and learn more about our offerings at our stall at #DES2024 #descriptivestats #inferentialstats #experientiallearning #dataanalytics #statistics

24

4 Comments

🔓 The data scientist is dead. Long live data science As generative AI rocks one industry after another, analytics services—which rely mostly on data—have felt the tremor. At the core of analytics is data science, and generative AI has automated the domain—over 75% of it. How exactly are India's largest analytics companies coping with generative AI? And who is the new data scientist? Read this story from our archives for free by signing up for The Ken's 30-day free trial. Click the link in the comments. ------------------ ✍ Seema Singh | March 2024 #TheKen #Data #DataScience #DataScientist #AI #GenAI #ArtificialIntelligence

5

1 Comment

Training a Machine Learning Model: A New Perspective 🤖 Training a machine learning model is like teaching someone to recognize animals 🐾. The concept of "epochs" in model training reflects how exposure to data impacts the model's learning process. 🕵️‍♂️ Model Trained for 1 Epoch: Similar to introducing a child to a tiger 🐅 for the first time, the model at this stage grasps only basic features. It struggles with intricate details due to limited exposure. 🐯 Model Trained for 100 Epochs: After multiple encounters with tiger images 🐯, the model can identify significant characteristics like stripes. However, it may still face challenges with noisy or unclear data. 🐅 Model Trained for 1000 Epochs: With 1000 epochs, the model masters even minor details, improving its pattern recognition abilities. Yet, overfitting risks increase, hindering its adaptability to new scenarios. ⚖️ The Challenge: Balancing the number of epochs is key for effective model training. Finding the sweet spot ensures the model learns optimally, avoiding underfitting or overfitting, resulting in a resilient and reliable model. #MachineLearning #DataScience #ModelTraining #ArtificialIntelligence

4

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

6

Join ETLHive's new data science batch and learn how to use data to solve real-world problems!👩‍💻 ☎️Call us on: Pune : 8055020011/ Mumbai : 7838539559 https://etlhive.com/ #datascience #datasciencecourse #datasciencejobs #datasciencetraining #datasciencelearning #dataanalytics #dataanalystjobs #newbatchstarting #newbatch

2

Join ETLHive's new data science batch and learn how to use data to solve real-world problems!👩‍💻 ☎️Call us on: Pune : 8055020011/ Mumbai : 7838539559 https://etlhive.com/ #datascience #datasciencecourse #datasciencejobs #datasciencetraining #datasciencelearning #dataanalytics #dataanalystjobs #newbatchstarting #newbatch

3

Ever wondered if R-squared in regression can be negative? It might seem odd—how can something squared be less than zero? But yes, under certain conditions, R-squared can go negative! Here’s how it works: R-squared (R²) tells us how well a regression model explains the variability in the data. It’s calculated like this: R² = 1 - (SSE / TSS) Where: • SSE (Sum of Squared Errors) shows how far off your model’s predictions are. • TSS (Total Sum of Squares) captures the total spread of the data from its mean. Usually, SSE is smaller than TSS, which keeps R² between 0 and 1. But if your model doesn’t have an intercept or fits the data poorly, SSE can be bigger than TSS, making R² negative. Why would that happen? In models without an intercept, the usual math breaks down, and the error terms can cause SSE to overshoot TSS. This leads to a negative R². So, if you see a negative R-squared, don’t panic—it’s just a sign that your model might need a tweak!

23

2 Comments

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

72

1 Comment

From racing dreams to data streams 🏎️💻: Meet Shivam, our Data Engineer who went from chasing the thrill of speed to crafting cutting-edge solutions at Anchanto. "As a child, I wanted to be a racer because the melody of the revving engines and speed has always given me thrills. And there is just something (that can’t be put into words exactly) about bikes and cars that excites me to the core. I joined Anchanto right out of college. Before entering the workforce, I was interested in research and solving the issues in processing Big Data. As a student, I even have my research published in reputed journals and have written conference papers. All of this led me to Anchanto. I joined right out of college and ever since Day 1, the opportunities to learn are endless. As a Software Developer at Anchanto, I primarily work on Anchanto Digital Shelf and Anchanto Control Tower. My responsibility includes leveraging machine learning techniques to solve product matching and clustering, content quality and sentiment analysis for enterprise brands. I work closely with Ayan, Varun, Nicholas, and the entire Data Product team. the Backend and ETL teams - it's an exciting environment. It helps that I have always wanted to do what I am doing right now (since a career in racing was not the best idea). What I love about working at Anchanto has been the amount of growth opportunities I have had in a span of just two years. My very first project was with Asendia, our valued customer. It gave me the opportunity to witness first-hand what our clients need and how to ensure the same is delivered. It has helped me learn and grow as an individual and become a better developer. From working on the backend and writing APIs to managing data pipelines and occasionally working on ETL - I can see how my skills and learnings have evolved during my time here. Apart from work, I'm passionate about bike and car rides, and trying different types of foods. My advice to juniors in this field is to be patient and keep trying different solutions until one works. The satisfaction and learnings from solving a problem yourself are invaluable. In closing, two quotes I try to live by: "Every day is a new beginning" and "For being a good speaker, you need to be a good listener first." These have shaped my attitude to life. I can't wait to see what's more in store." #MeetTheteam #Anchanters #TeamCulture #CompanyCulture #Growth #DigitalShelf #SaaS

50

1 Comment

How much correlation implies causation Granted that domains like Marketing abuse correlation. You will come across correlation coefficient depicted proudly in Marketing research papers, whitepapers, blogs etc. as though they are causal !! But the over correction for it i.e. nonchalantly uttering 'Correlation is not causation' is also equally harmful. Many use this phrase to dismiss arguments without thinking about them thoroughly. Such callous usage of the phrase leads to outright rejection of any further discussion into causality. Nobody is saying "correlation is causation" (at least not right off the bat). But a strong correlation does lead to the question of causality. Correlation at the least expresses a hypothesis that the two variables or phenomenon might be related. It also leads us to explore more robust methodology like Hill's criteria to establish causality . I think the best summary to all this is by Randall Munroe XKCD cartoon. "Correlation doesn't imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing 'look over there'".

97

7 Comments

Srikanth invites you to step into the future with Fractal's AI Leadership Specialization, 'Leadership Strategies for AI and Generative AI.' This specialization, designed by Fractal's experts, is perfect for leaders and consultants in the AI field. Discover how to leverage generative AI across various business domains. Master the art of integrating AI into your operations with a strong focus on data privacy and ethics. Equip yourself with the skills to set impactful KPIs, interpret data insights, and enhance your generative AI initiatives. You will learn to: 💠Evaluate potential applications of generative AI across diverse business domains. 💠Integrate generative AI into business operations effectively, considering data privacy and ethical implications. 💠Establish KPIs, interpret data-driven insights, and devise optimization strategies for generative AI initiatives. This specialization is available exclusively on Coursera: https://lnkd.in/dkbTz9fC Enroll today and transform your career with Fractal's AI Leadership Specialization. #AISpecialization #FractalCourses #Coursera

331

4 Comments

📊 Machine Learning Landscape: Comparisons & Insights Below shows comprehensive diagram of machine learning techniques, and it sparked some interesting comparisons. Let's dive deeper! 🔍 Key Areas & Comparisons: Supervised vs. Unsupervised Learning: • Supervised: Requires labeled data, great for prediction tasks • Unsupervised: Works with unlabeled data, excels at pattern discovery Comparison: Supervised often yields more accurate results, but unsupervised can uncover hidden patterns we didn't know to look for. Classification vs. Regression: • Classification (e.g., Logistic Regression, SVM): Predicts categories • Regression (e.g., Linear, Polynomial): Predicts continuous values Comparison: Choose classification for discrete outcomes (like spam detection) and regression for continuous outcomes (like price prediction). Tree-based vs. Neural Network approaches: • Tree-based (e.g., Random Forest): Interpretable, handles non-linear data well • Neural Networks: Powerful for complex patterns, especially in large datasets Comparison: Trees are often easier to explain to stakeholders, while neural networks can capture more intricate relationships in data. Clustering Algorithms: • K-Means: Fast, works well with spherical clusters • DBSCAN: Handles clusters of different shapes, doesn't require pre-defined number of clusters Comparison: K-Means is simpler and faster, but DBSCAN is more flexible for complex data distributions. Dimensionality Reduction: • PCA: Linear technique, preserves global structure • t-SNE: Non-linear, better for visualizing high-dimensional data Comparison: PCA is faster and works well for many applications, while t-SNE often produces more visually appealing results for complex datasets. Ensemble Methods: • Bagging (e.g., Random Forest): Reduces variance, good for complex datasets • Boosting (e.g., XGBoost): Reduces bias and variance, often achieves higher accuracy Comparison: Bagging is less prone to overfitting, while boosting can produce more accurate models but requires careful tuning. Deep Learning vs. Traditional ML: • Deep Learning: Excels with large datasets, automatic feature extraction • Traditional ML: Often works well with smaller datasets, requires feature engineering Comparison: Deep learning shines in areas like computer vision and NLP, while traditional ML is often sufficient (and faster to train) for structured data problems. 💡 Pro Tip: The best algorithm often depends on your specific problem, dataset, and constraints. Experimentation is key!

41

1 Comment

Tiger Analytics partners with us to digitize and streamline data across forecasting, production, distribution, and programmes operations—boosting efficiency to better serve our beneficiaries 🤝 #TigerAnalyticsXAkshayaChaitanya

60

Who is Srikanth Velamakanni, the founder of Fractal Analytics? Srikanth Velamakanni is a prominent figure in the world of data analytics, known for his innovative approach to business intelligence. As the founder of Fractal Analytics, he has revolutionized the industry with his cutting-edge solutions and strategic insights. See here - https://lnkd.in/e67CWmDN #SrikanthVelamakanni #FractalAnalytics #DataAnalytics #BusinessIntelligence #Innovator #TechLeader #AI #BigData #DataScience #Innovation #TechIndustry

3

📊 How Exit Polls Work: A Data Science and Statistical Perspective 🗳️ Exit polls are a fascinating blend of data science and statistics, offering a preliminary glimpse into election outcomes. Here's a breakdown of how they work and why they're so significant, especially in the context of the 2024 Indian General Elections: 🔍 Design and Sampling: Uses random and stratified sampling to ensure diverse voter representation. Polling stations are chosen to capture urban-rural divides and various demographic groups. 📋 Data Collection: Trained surveyors collect data from voters exiting polling stations. Standardized questionnaires gather voting choices and demographic details. 📈 Data Analysis: Responses are weighted to correct sampling biases. Statistical models like regression analysis and machine learning identify trends and predict outcomes. 🔮 Predictive Analytics: Historical data and machine learning enhance prediction accuracy. Trends such as urban vs. rural voting patterns, youth voter preferences, and economic issue impacts are analyzed. 📰 Interpretation and Reporting: Results are reported with confidence intervals and visualized through graphs and heat maps for clarity. Why Exit Polls Matter: Real-time Insight: Provides immediate insights into voter behavior. Strategic Decisions: Helps political parties understand voter sentiment. Public Engagement: Engages the public and media, fostering participation and anticipation. Conclusion Exit polls, backed by robust data science and statistical methods, are vital for understanding voter behavior and predicting election trends. Let's dive into the data and uncover the stories behind the numbers! 📊🗳️ #DataScience #Statistics #ExitPolls #Elections2024 #India #PredictiveAnalytics #VoterBehavior #MachineLearning #PoliticalAnalysis #SurveyMethodology #ElectoralStudies

23

1 Comment