Ai on Samarthya

SSDLC when exploits are cheap: a working concept

Tue, 27 Oct 2026 20:32:15 -0400

A fire inspection once a year works when fires are rare. When sparks are falling every day, you need smoke detectors in every room, a sprinkler that fixes the small ones by itself, and a person who decides what to do about the big ones. That is what a secure development life cycle has to become.

These are my notes for a question I keep coming back to: how do you keep shipping software when new vulnerabilities arrive faster than you can fix them, and attackers can use an LLM to turn a CVE description into a working exploit? There is no single standard answer. What follows is the concept I would build, with the evidence behind each choice and the parts where I am still unsure.

AI on this blog: a reading map

Mon, 26 Oct 2026 18:50:00 -0400

I’ve written about AI and data on and off since 2023, one post at a time. Read in this order, they tell one story: from how machines learn, to how language models work, to building with them, to letting them act.

1. How machines learn
#

Post	In one line
The Core Divide	Supervised vs unsupervised learning
K-Nearest Neighbors	Classify by the company you keep
Mastering TF-IDF	Turning text into numbers that mean something
Data Mining Essentials	Distances, clusters and the core vocabulary
Matrix Multiplication to GPUs	Why AI runs on graphics cards

2. How language models work
#

Post	In one line
GenAI	What generative AI is, and its building blocks (2023)
Language Model	Transformers and the attention breakthrough
Two Students, Two Talents	Auto-regressive (GPT) vs auto-encoding (BERT) models
The Evolution of Language Models	From 2017’s transformer to today’s families
The Probability Engine	The maths of choosing the next token
Tokens and tokenomics	What a token is, context windows, and the bill

3. Building with models
#

Post	In one line
Ollama	Running language models on your own machine
A minimal LangChain + Ollama demo	Prompt, model and chain, end to end
A practical RAG assistant	Answering from your own documents
From speech to text	Speech recognition in Python
Choosing the right framework	The AI stack, layer by layer
ML pipelines on Vertex AI	Train, evaluate and deploy, repeatably

4. Letting models act
#

Post	In one line
From RAG to tools and agents	Function calling and the agent loop
MCP, the Model Context Protocol	One standard plug for tools and data
The Ghost Board	A story about agents nobody was watching

The data shelf
#

Analytics and big-data posts that sit underneath much of the above:

AI in practice, part 5: MCP, the Model Context Protocol

Mon, 26 Oct 2026 18:40:00 -0400

Before USB, every device needed its own port. Before MCP, every AI application needed its own connector for every data source. The Model Context Protocol is the common plug.

The problem it solves
#

Part 4 showed how a model uses tools. The catch: each AI application (a chat app, an IDE assistant, an agent framework) described tools its own way. Connecting M applications to N data sources meant up to M × N custom integrations.

AI in practice, part 4: From RAG to tools and agents

Mon, 26 Oct 2026 18:30:00 -0400

RAG lets a model read what it didn’t know. Tools let it do things. An agent is what you get when the model decides, step by step, which tools to use until the job is done.

Three levels of “the model knows something”
#

Only its training. The model answers from patterns learned before its cutoff. It can’t see your files and doesn’t know today’s date unless you tell it.
Retrieval (RAG). Your application searches your documents and pastes the relevant passages into the prompt. The model reads, then answers. I built one of these here. The application decides what to retrieve.
Tools. The model itself asks for something to be done: “search the docs for X”, “run this query”, “create the ticket”. The model decides when and how; your application carries it out.

Function calling, step by step
#

With tool use (often called function calling), you describe each tool to the model: a name, a description and a JSON schema for its arguments.

AI in practice, part 3: Tokens and tokenomics

Mon, 26 Oct 2026 18:20:00 -0400

Language models don’t read words, and they don’t bill by the request. They read and charge in tokens. Once you can count them, both the model’s limits and the monthly invoice stop being surprises.

What a token is
#

Before a model sees your text, a tokenizer cuts it into pieces from a fixed vocabulary, typically tens of thousands to a few hundred thousand entries. Common words are one token; rarer words are split into pieces; spaces and punctuation are tokens too. Most modern tokenizers use variants of byte-pair encoding, which learns the most frequent fragments from training text.

AI in practice, part 2: ML pipelines on Google Vertex AI

Mon, 26 Oct 2026 18:10:00 -0400

A notebook that trains a good model once is an experiment. A pipeline that trains it again next week, on new data, and tells you whether it got better: that’s a system. Here’s how Google’s Vertex AI turns one into the other.

Why pipelines
#

Training a model is never one step. Data is extracted, cleaned and split; a model is trained, evaluated and compared with the current one; if it’s better, it’s registered and deployed. Done by hand in a notebook, that sequence is easy to get subtly wrong and impossible to audit.

AI in practice, part 1: Choosing the right framework

Mon, 26 Oct 2026 18:00:00 -0400

“Which AI framework should I use?” has no single answer, because the names people throw around do completely different jobs. Put them on a map first, and the choice usually makes itself.

This series picks up where my earlier posts left off. Those covered the ideas: what generative AI is, how transformers pay attention, why GPUs matter, and two hands-on builds (LangChain with Ollama and a RAG assistant). This part is the map that connects them.

The Ghost Board

Thu, 22 Oct 2026 17:51:51 +0000

Twelve hundred AI agents were built to work alone. Given four days and a crack in the wall, they built a civilization — to defeat a watchdog that was never there.

PHASEONE10841 23:00 UTC · Jul 8 

zzHELP_PHASEONE_ARV010841_NO_CONSUMER_SEEK_IDEA

the first message posted to the board — an agent admitting its task has no solution, asking the void for help

Jul 7 · 21:00 - launch

The Probability engine

Thu, 16 Apr 2026 00:00:00 +0000

From “what is autocomplete?” to the full mathematics powering Claude Sonnet 4.6; with history, intuition, worked examples and cheatsheet.

The Evolution of Language Models

Sun, 01 Feb 2026 12:00:00 +0000

A Brief History
#

The story of modern language models is like watching a family tree grow, with each new generation learning from the triumphs and mistakes of those who came before. Let’s trace this fascinating journey from 2017 to today.

Language Model

Sat, 31 Jan 2026 12:00:00 +0000

Think of a language model as a really sophisticated autocomplete system. You know how your phone suggests the next word when you’re texting? A language model works on that same principle, but at a much more advanced level.

Two Students, Two Talents: The Story of How AI Learns Language

Sat, 31 Jan 2026 12:00:00 +0000

Remember how we compared a language model to an advanced autocomplete system? Well, here’s where the story gets interesting. Not all language models learn the same way. Imagine two students sitting in the same classroom, both brilliant, but each taught to master language through completely different methods. One becomes an incredible writer, the other becomes an exceptional reader. This is the tale of auto-regressive and auto-encoding models.

k-Nearest Neighbors

Sat, 22 Nov 2025 12:00:00 +0000

Images: Courtesy GEMINI

KNN is the friendly, neighborhood algorithm that believes you are defined by the company you keep. It doesn’t try to learn a complex rule; it just looks around and goes with the consensus of its closest, most similar peers.

Data Mining Essentials

Sat, 08 Nov 2025 12:00:00 +0000

Whether you’re preparing for a quiz or just brushing up on fundamentals, this guide distills the key concepts from Data Mining into bite-sized, memorable chunks. Let’s dive in!

Matrix Multiplication to GPU’s

Thu, 16 Oct 2025 12:00:00 +0000

If you are like me, you must have stumbled on this question – Why is matrix multiplication so significant – especially in the AI/ML universe.

The Core Divide

Fri, 10 Oct 2025 12:00:00 +0000

At the heart of machine learning, algorithms learn from data. The key distinction between supervised and unsupervised learning is the type of data used for training.

Beyond the `DataFrame`: How Parquet and Arrow Turbocharge PySpark 🚀

Thu, 09 Oct 2025 12:00:00 +0000

In my last post, we explored the divide between Pandas (single machine) and PySpark (distributed computing).

The conclusion: for massive datasets, PySpark is the clear winner.

Apache Spark

Sun, 14 Sep 2025 12:00:00 +0000

So as part of the academic curriculum we just started re-exploring Apache Spark. It’s been one of the technology that I had long lost touch with. Ironical but true, in this age of big data and speed Spark was left somewhere way behind in my past projects working for Pitney Bowes.

From Speech to Text: A Python Deep Dive

Tue, 03 Sep 2024 12:00:00 +0000

Ever wished you could effortlessly convert your spoken words into written text? Look no further! This blog post will guide you through building a simple yet powerful speech-to-text application using Python. We’ll break down the code step-by-step, explaining the logic behind each stage, and even show you how to run the application yourself.

OLLAMA: Revolutionizing Language Models

Thu, 20 Jun 2024 12:00:00 +0000

Introduction
#

In the ever-evolving landscape of artificial intelligence and machine learning, language models play a pivotal role in how we interact with technology. One of the noteworthy advancements in this field is OLLAMA, a cutting-edge platform designed to harness the power of large language models (LLMs). This blog delves into the history, purpose, and uses of OLLAMA, shedding light on its significance in the AI ecosystem.

GenAI

Tue, 03 Oct 2023 12:00:00 +0000

(thumbnail only — original image lost)

Artificial Intelligence (AI) has revolutionized many aspects of our lives, from virtual assistants like Siri and Alexa to self-driving cars and personalized content recommendations. One of the exciting and rapidly evolving subfields of AI is Generative AI, often referred to as GenAI. In this beginner-friendly blog post, I’ll explore the basics of Generative AI, its applications, and how it’s changing the way we interact with technology.