The notebook
Essays and notes, newest first. Follow a topic or read through a series.
Browse topicsRead a seriesYearly archiveSearch the notebook
The September 2026 Tata Sons dispute shows why ownership, board seats and voting rights need separate edges. Public filings reveal shared boards, corporate continuity and even relationships that should be removed from a graph.
The graph can connect verified boards with bank health, company fundamentals, ownership signals, auditors, and subsidiaries—but it cannot turn aggregate data into borrower-level lending exposure.
A verified snapshot of all 100 NIFTY 100 boards turns corporate governance into a network: 954 current appointments, 33 multi-board directors, and 17 company pairs sharing at least two directors.
A reproducible P/BV and excess-returns framework for lenders: perpetual growth, finite fade, recovery assumptions, and the difference between creditor recovery and equity value.
Asset-liability mismatch, refinancing concentration and credit stress can interact in lenders. A framework for reading the IL&FS and DHFL cases, interpreting liquidity buffers and keeping creditor recovery separate from equity value.
A lender’s Cost-to-Income Ratio tells you how efficiently the machine runs. It says nothing about whether the machine’s output actually arrives as cash. This post connects operating efficiency to DCF valuation through the lens of...
A first-principles walkthrough of cash-flow discounting, CAPM, WACC, terminal value and sensitivity analysis, with explicit illustrative inputs and reproducible arithmetic.
How discounting connects home-loan EMIs, lender funding, retained capital and growth decisions, with explicit illustrative assumptions rather than unsourced issuer price targets.
Asset intelligence is the real payoff of the AI cycle, but only if quality capture and infrastructure inequality are addressed head on.
The winners in AI will not only own models, they will own power, silicon, and the capital stack that sustains iteration speed.
When AI makes the first draft cheap, the winners are the teams that can prove quality, security, and reliability at speed.
A clear-eyed look at why this AI cycle feels bubbly, why it is still anchored in real economics, and why software creation is being commoditized faster than most teams are ready for.
Automotive OEM Restructuring for Data Monetization and Innovation - Part 3 In Part 1 of this series, we explored how automakers globally — from Volkswagen and Mercedes-Benz to Tata Motors and Mahindra — spent the...
Automotive OEM Restructuring for Data Monetization and Innovation - Part 2 In Part 1 of this series, we explored how automakers globally — from Volkswagen and Mercedes-Benz to Tata Motors and Mahindra — spent the...
Automotive OEM Restructuring for Data Monetization and Innovation - Part 1 Over the past decade, major automakers have radically reorganized their businesses to transition from hardware-centric manufacturers into software-driven mobility companies. Between 2015 and 2025,...
High-Cost R&D Domains: Crash Tests, Batteries, Autonomous Validation, Homologation Modern vehicles are incredibly complex to develop, and some aspects of R&D incur especially high costs.The most expensive components of automotive R&D in 2025 include safety/crash...
Global Automotive Industry 2025 Overview The worldwide automotive industry rebounded strongly from the pandemic, with 2024 global motor vehicle production reaching 92.5 million units, just slightly (−1%) below 2023’s level. This volume is on par...
Autonomous Decision-Making Systems: The Next Frontier in AI and Data Monetization Autonomous decision-making systems are AI-driven platforms that can make and execute decisions without human intervention – spanning use cases from automated financial trading to...
Introduction: What Is Hyperpersonalization? Hyperpersonalization is the use of real-time data and AI to deliver uniquely tailored experiences to individual users, far beyond traditional customer segmentation. Unlike conventional personalization that might group users by age...
Amazon S3: From Simple Storage to Platform Monetization Engine (2006–2025) [Part -3] Part-1 Part-2 Part-3 6. Cyclic Pressures and Strategic Resilience Over nearly two decades, AWS S3 has weathered multiple economic cycles – from recessions...
Amazon S3: From Simple Storage to Platform Monetization Engine (2006–2025) [Part 2] Part-1 Part-2 Part-3 4. Business Model Evolution and S3’s Role Amazon S3 did more than make money from storage – it fundamentally altered...
Amazon S3: From Simple Storage to Platform Monetization Engine (2006–2025) Part-1 Part-2 Part-3 Sources 1. Genesis and Strategic Intent Amazon Web Services (AWS) launched Amazon S3 (Simple Storage Service) in March 2006 as one of...
Amazon S3: From Simple Storage to Platform Monetization Engine (2006–2025) [Sources & References] Part-1 Part-2 Part-3 Sources: Sourced content includes SEC filings and Amazon investor reports (for financial milestones), AWS blog posts and books (for...
Linear regression models rely on a loss function to quantify how far predicted values are from the actual observations. Minimizing this loss is what drives the learning process. In practice there are several ways to...
Linear regression offers a simple way to relate a numeric target with one or more features. Three terms often appear together when evaluating how strong that relation is:
In real-world datasets, imbalanced class distributions are more common than balanced ones. Simply shuffling and splitting data may lead to training and test sets that don’t preserve the original label proportions. Enter Scikit-Learn’s StratifiedShuffleSplit —...
When preparing data for machine learning or statistical analysis, you often need to transform continuous variables into categorical bins. This is where pandas.cut becomes invaluable.
In the world of machine learning, how you split your data is just as important as the model you train. The widely used train_test_split function from Scikit-Learn is deceptively simple, yet foundational. What’s more interesting...
Java 8’s Stream API didn’t just introduce a functional syntax — it quietly redefined how developers could leverage multicore processors. With parallel streams, Java brought a high-level abstraction for concurrent computation, underpinned by the Fork/Join...
Java 8 introduced the Stream API, a major shift toward functional-style programming in Java. Among the key design elements that power this shift are the functional interfaces like Consumer, Supplier, and Function. While the Stream...
NumPy is often described as the foundation of the scientific Python ecosystem. But beyond performance and vectorization, what makes NumPy truly enduring is its clean, minimal, and orthogonal API design , shaped over years by...
Pandas is well-known for its intuitive and expressive API. At the core of this usability lies the thoughtful design of three main abstractions: Index, Series, and DataFrame. Each of these was introduced with deliberate consistency...
Revised 2 October 2026: Corrected per-column dtype inference and the manager display; removed unsupported benchmark timings and distinguished later APIs from the historical example. Original publication date retained.
Kubernetes and the Cloud-Native Revolution
Understanding Linux cgroups: The Foundation of Resource Isolation in LXC
Normalization is a foundational concept in relational database design. It governs how data is structured, stored, and maintained to ensure consistency, minimize redundancy, and optimize performance under specific access patterns. While there are many normal...
Meeting Basel III regulatory requirements is not just a matter of compliance — it is an engineering challenge. Global banks operate hundreds of systems across risk, finance, treasury, trading, and operations. Generating Basel III reports...
While capital adequacy addresses a bank’s ability to absorb losses, Basel III also places significant emphasis on liquidity management. Liquidity risk — the inability to meet short-term obligations — played a central role in the...
Under Basel III, capital adequacy is a cornerstone of prudent banking regulation. It ensures that banks have enough capital to absorb losses, withstand financial stress, and continue operations without threatening the broader financial system. Capital...
Basel III introduces a set of comprehensive reforms to strengthen regulation, supervision, and risk management within the banking sector. Global banks are required to meet stringent capital adequacy, leverage, and liquidity requirements — all backed...
The paper “All File Systems Are Not Created Equal: On the Complexity of Crafting Crash-Consistent Applications” by Pillai et al., presented at OSDI 2014, delivers a sobering insight: writing crash-consistent applications is far more complex...
Traditional sales often focus on matching products to requirements. But in the world of complex systems and enterprise solutions, customers rarely articulate their deepest problems clearly. This is where solution selling stands apart — it’s...
Docker is transforming how developers build, ship, and run software. At its core, Docker provides a lightweight containerization platform that simplifies dependency management and enables the “build once, run anywhere” promise — especially in the...
As Hadoop-based data lakes grow in volume and variety, data governance becomes mission-critical. Governance is not just about compliance — it’s about trust, transparency, and control. Without it, a data lake risks turning into a...
As Hadoop matures into an enterprise-grade data platform, security becomes more than a checkbox — it is foundational. Among the many pillars of Hadoop security, Kerberos-based authentication stands out as the linchpin that enables trusted...
In the evolving landscape of cloud infrastructure, Amazon S3 (Simple Storage Service) emerges as the gold standard for object storage. Launched in 2006, S3 is not just a scalable storage system — it is a...
As the ecosystem of distributed data processing evolves, two major frameworks — Apache Spark and Apache Beam — emerge with distinct approaches. Both aim to simplify and scale data processing across clusters, but they cater...
Apache Spark Streaming brings the power of Spark’s batch processing engine into the world of real-time data. Built on top of the core Spark engine, it allows developers to process live data streams in a...
Apache Spark 1.4, one of the most widely adopted distributed data processing engines of its time, leans heavily on Akka — the actor-based toolkit — for its internal messaging, coordination, and fault tolerance. Before Spark...
In the world of Scala, purely functional programming offers a powerful way to write predictable, composable, and reusable code. Libraries like Cats and Scalaz bring abstractions from category theory and algebraic structures to mainstream developers,...
The Java Virtual Machine (JVM) has always been a powerful abstraction layer, and its memory management system — particularly garbage collection (GC) — is a critical part of its performance story. As of mid-2016, the...
In Scala, immutability is not just a style — it’s a principle that shapes how we write correct, composable, and predictable code. Immutable types prevent whole classes of bugs at compile time, enable safe concurrency,...
When we use Git every day, it feels simple: git commit, git push, git merge. But underneath the CLI, Git is one of the most interesting and elegant distributed systems in active use today. Unlike...
It’s June 30, 2016, and building distributed systems still feels more like an art than an exact science. Despite advances in consensus algorithms, databases, and infrastructure tooling, ensuring correctness under failure remains one of the...
As of June 22, 2016, Apache Hadoop YARN (Yet Another Resource Negotiator) continues to power many of the world’s largest data platforms. At the heart of YARN lies the Resource Manager (RM) — a global...
As of June 18, 2016, Apache Kafka continues to dominate the world of high-throughput messaging systems. One of the lesser-known but critical features contributing to its performance is zero-copy, a Linux kernel optimization that Kafka...
It’s June 2016, and logistic regression remains one of the most reliable, interpretable models for binary classification problems. With SparkML and Scala, it’s now easier than ever to apply this powerful technique to large-scale datasets...
It’s June 2016, and while machine learning continues to evolve with deep neural nets and complex ensemble methods, linear regression remains one of the most trusted tools in the data scientist’s toolbox.
Revised 2 October 2026: Corrected next-offset semantics, crash behavior, transaction boundaries and retention claims while retaining the Kafka 0.9-era context. Original publication date retained.
By May 2016, Apache HBase had firmly established itself as a go-to NoSQL database for low-latency, high-volume, and sparse data use cases. But what makes HBase interesting — and often misunderstood — is how it...
By May 2016, SparkML had emerged as a practical and scalable way to build and deploy machine learning pipelines directly on distributed infrastructure — and with YARN as the resource backbone, operationalizing ML at scale...
By April 2016, YARN had already become the cornerstone of resource management in the Hadoop ecosystem — powering everything from MapReduce to Spark, Hive, and Tez. But even as YARN matured, the data infrastructure landscape...
As of April 2016, the need for scalable and resilient systems has made actor-based concurrency more relevant than ever. In the JVM world, Akka brings this model to life — drawing inspiration from Erlang’s telecom...
As of April 2016, building concurrent systems is no longer optional — it’s a necessity. Whether you’re writing backend services, data processing pipelines, or reactive systems, safe and scalable concurrency is foundational.
As of March 2016, Apache Oozie remains one of the foundational workflow engines in the Hadoop ecosystem. Designed for orchestrating complex, multi-stage data processing pipelines on HDFS and YARN, Oozie helps define, schedule, and monitor...
As of March 2016, large financial institutions are under constant pressure to provide timely, accurate, and auditable regulatory reports. With diverse, siloed systems across geographies, the question is no longer if data centralization is needed...
By late February 2016, Apache HBase had become a staple of the NoSQL world — the go-to system when teams needed low-latency, high-throughput access to massive, sparse, and semi-structured datasets.
As of February 2016, HDFS (Hadoop Distributed File System) continues to be the foundational layer for most big data platforms, including Spark, Hive, Tez, HBase, and more.
By the start of 2016, Apache Kafka had cemented its position as a core infrastructure layer for real-time data movement. At its heart was a deceptively simple idea: decouple producers and consumers, and let the...
In January 2016, bytecode manipulation isn’t just a niche trick — it’s a powerful capability that lets you inspect, modify, and enhance Java or Scala applications without altering source code.
It’s January 2016, and the field of data engineering is almost unrecognizable from where it stood just a few years ago.
As 2015 draws to a close, Haskell continues to quietly influence how we think about abstractions, structure, and semantics in computation — not just locally, but distributedly.
In December 2015, as Spark 1.6 was gaining traction across the data world, one of its most powerful ideas was hiding in plain sight: homomorphism.
Revised 2 October 2026: Removed the truncated closing suggestion and supplied a modest conclusion consistent with the existing discussion. Original publication date retained.
In the landscape of programming languages in 2015, Haskell continues to quietly demonstrate how radical ideas like infinite data structures, tail recursion, and lazy semantics can fundamentally reshape how we express algorithms — especially those...
By October 2015, Apache Hive had firmly established itself as the SQL engine of choice in the Hadoop ecosystem — powering data warehouse-style workloads at scale.
Revised 2 October 2026: Replaced the truncated final sentence with a modest conclusion consistent with the article; historical product guidance was not modernized. Original publication date retained.
By October 2015, Apache Spark had officially moved from being a promising successor to MapReduce into the mainstream engine of choice for many big data workloads.
By September 2015, Hadoop was no longer a fringe technology. It had become the de facto platform for big data infrastructure — and at the heart of this evolution was a powerful but often misunderstood...
In 2015, a quiet but massive transformation is underway — the rise of data engineering as a first-class discipline.
In 2015, search infrastructure was undergoing a quiet revolution.
When I first started tuning JVM-based applications at scale, the prevailing belief was: “The JVM is fast enough. Let it handle the rest.” And for the most part, that worked — the JVM is a...
When I first began working with Oracle on massive-scale transactional workloads, performance tuning was more art than science. You’d hear things like “The optimizer is your friend” — which often felt ironic when queries took...
When I first began building domain-specific languages (DSLs), I focused on surface-level structure. Syntax. Keywords. Aesthetics. I believed if the language looked clean and intuitive, people would adopt it. And to an extent, that worked...