Build logs from working on government tech, edge ML, and defence simulations. Browse by tag RSS ↗
A case study of something I actually shipped: a tool that lets an officer ask a question in plain words and get the right section of the new BNS, BNSS, and BSA, explained simply. Here is how it works and why I kept it boring.
After the Bhashini posts, a reader asked for the same treatment of PM GatiShakti. It is a very different kind of platform: not AI models, but one shared map for planning India's infrastructure. A simple, fact-checked explainer.
You do not need an expensive GPU server to do real, useful AI. A practical, honest look at what a single modest machine with no GPU can run, what it cannot, and how I host a whole platform on one small box.
Recursion looks scary but it is just a function that calls itself on a smaller piece, and the call stack does the bookkeeping. It is the natural tool for trees, the shape a document like a manual actually has.
Bhashini lets you speak in one Indian language and be understood in another. People assume it is one giant AI. It is not. A simple, fact-checked look at the real models behind it, the base model for translation, and how they were trained.
The two simplest ways to decide what gets handled next: a queue (first in, first out) and a stack (last in, first out). Where each one shows up in real systems I have built, with examples.
The 'halve it every time' idea. How binary search finds something in a million items in about twenty steps, where it quietly powers databases and even git, and the one condition it needs.
Large language models will confidently invent a law section that does not exist. Retrieval-Augmented Generation (RAG) is the fix. A plain-English explainer, grounded in a real legal-lookup tool I built for police.
The single most useful data structure I reach for: the hash map. What it is, why lookups are instant, where I use it in real systems, and the catches to know about. No jargon.
Forget the hype. A grounded, plain-English guide to how the boring, explainable machine learning models can help a real investigating officer sort a pile of cases, spot the odd one out, and defend the reasoning, without ever pretending to be a detective.
Everyone talks about neural networks and LLMs. But the models quietly running your bank, your fraud alerts, and half of Kaggle are older, simpler, and far more boring. A plain-English tour of the real workhorses.
A plain-English guide to how AI, machine learning, neural networks, and large language models actually relate. They're not competing ideas — they're nested inside each other. Written by an ex-Army officer who now builds them.
A lockdown build: reusing an old RO water cabinet and its solenoid valve, an Arduino Nano, and an ultrasonic sensor to dispense sanitizer touch-free without false triggers.
A 2.1 sound system built after a proper study of subwoofer acoustics: a ported MDF enclosure sized from Thiele-Small parameters, an NE5532 pre-amp, and a TDA2050 power stage for deep, punchy bass.
Building a full electronic drum kit from piezo triggers and a DIY rack, converting strikes to MIDI, and driving the sounds from GarageBand on an iPhone with headphones.
An IoT build that pulls live COVID-19 figures for India from a web API on an ESP8266 and scrolls them across a 64x32 P4 LED matrix using the PxMatrix library.
Inter-Integrated Circuit (I2C): how the master/slave two-wire bus works, where it is used, and a simple Arduino example for reading a sensor over I2C.
How a two-axis analog joystick works: two potentiometers for X and Y plus a push switch, its pinout, and where joysticks are used from games to surgical robots.
A 4x4x4 LED cube driven by an Arduino UNO and 64 LEDs, with the layer-and-column multiplexing code that animates it.
A from-scratch Marshall-inspired Bluetooth speaker in a Deodar wood box, with a TDA8932 amp board, JBL drivers, and a hand-made golden jute grill. The full bill of materials and build.
The MPU6050 motion sensor: a 3-axis accelerometer plus a 3-axis gyroscope with an onboard Digital Motion Processor. What it measures, where it is used, and an Arduino example to read it over I2C.
A compact, space-saving octopad built from piezo sensors and aluminium-composite triggers, wired into an existing Roland TD-11 module for silent, headphone practice.
A from-scratch build log of a man-packable surveillance quadcopter on a Pixhawk flight controller, including every part, why it was chosen, and the trial and error that got it flying.
Serial Peripheral Interface (SPI): the four wires (MOSI, MISO, SCK, CS), how a transaction works, its pros and cons versus I2C, and a simple Arduino example.
How ultrasonic sensors measure distance with sound, the parts of the common HC-SR04 module, and an Arduino example that classifies an object as Near, Far, or Too Far.
Air drums that trigger MIDI notes when your hands pass over infrared sensors, built on an ATmega8 at the MIDI baud rate, with a 555-based 38kHz IR front end and TSOP1738 receivers.
A compact desktop speaker built from leftover teak wood and a PAM8610 amp, with a 4 inch driver and a 1.5 inch driver, sanded and polished into a tidy little cube for laptop audio.