Blog

Wavelets & Clustering to Detect Odd vs. Even Numbers

A deep dive into how wavelet transforms and clustering can uncover symbolic structure (parity) from raw numerical sequences. Includes visualizations, failure cases, and thoughts on feature engineering in the LLM era.

Applied Linear Regression Projects (x5)

Five different regression projects in Python (salary, cars, wine, insurance, trips). Focus on data prep, model selection, evaluation metrics, and practical trade-offs rather than pure theory.

Comparing ML Algorithms: KNN vs SVM vs Decision Tree

Decision Tree, SVM, and KNN on the same dataset, compared with accuracy, precision, recall, and F1. Shows how each model behaves, where it fails, and how to interpret metric trade-offs.

Text Generation with Q-Learning

A character-level Q-Learning setup that tries to reach target sentences. Explains states, actions, rewards and how RL can be used for toy NLP tasks without any deep learning.

Building JumpNet — A Vision-based Agent for a One-Button Game

How I recorded gameplay data, designed the training pipeline, and evaluated a CNN-based agent that decides when (and how long) to jump. Covers data collection pitfalls, modeling strategy and real-time GUI.

Series (3 parts):

• Part 1 — Data foundation
• Part 2 — Training the policy
• Part 3 — Real-time inference

Face Landmarks Detection with OpenCV DNN + Facemark

Real-time 68-point facial landmark tracker with EMA smoothing and CSV logging. Uses OpenCV DNN for detection and Facemark LBF for landmark prediction, plus temporal smoothing for stable overlays.

Modular Snip Recorder / Advanced Dataset Viewer

A modular data collection tool that records screen + keypresses and lets you browse and inspect behavior cloning datasets afterwards. Designed to debug agents like JumpNet before training.

Series (2 parts):

• Part 1 — Tool design
• Part 2 — Viewer & insights

Real-Time Image Color Palette Extractor

A Streamlit app that extracts dominant colors from images in real time using k-means, CIE LAB, and ΔE2000. Includes details on color distance, clustering stability and UX choices.

VAE vs VQ-VAE for Compact Face Representations

An experiment-heavy comparison between continuous and discrete latent spaces for compressing faces. Architecture details, training curves, and case studies where each method succeeds or fails.

Conditional GAN — Sketch Generation (Berlin Sketches)

A conditional GAN trained on black-and-white Berlin sketches, generating class-specific images from labels. Walkthrough of training tricks, failure modes, and what made convergence hard in practice.

Building an Edge AI Sound Classifier on Raspberry Pi Pico

End-to-end TinyML pipeline on RP2040 classifying audio into baby cry, doorbell, smoke alarm, and background. From dataset collection to feature extraction and C++ firmware with Logistic Regression on-device.

Series (3 parts):

• Part 1 — Data collection
• Part 2 — Features & training
• Part 3 — Firmware & demo

Edge AI Trend Alarm on Raspberry Pi Pico

Edge AI pipeline detecting workload-induced heating trends on the Pico. Data logging, rolling OLS slopes, Logistic Regression, and real-time inference in C++ with oversampled ADC and hysteresis FSM.

Series (3 parts):

• Part 1 — Data gathering
• Part 2 — Training
• Part 3 — Firmware integration

Building a Mini SCADA Console on Raspberry Pi Pico (C++)

USB-CDC based command console simulating an industrial field device: telemetry (internal temp via ADC4), logging and LED control. Shows how to structure firmware for simple SCADA-style interactions.

Local PDF QA with Lightweight LLMs

Building a small, local-first PDF summarizer and QA system. Focus on chunking strategies, retrieval quality, speed, and keeping everything private on a normal laptop.

How Do You Actually Optimize Agents?

A systems-level look at agent optimization beyond prompt tuning. Why most agent failures are task design failures — and how constraints, feedback loops, and human supervision shape reliable agent-in-the-loop systems.

My First Public ML Talk: Human-in-the-Loop → Agent-in-the-Loop

A short recap of preparing and delivering my first public ML talk, focused on transitioning from human-in-the-loop systems to agent-in-the-loop architectures. Includes reflections, structure, and what I’d improve next time.

How I Built a Simple Shell in C — System Programming (1/3–3/3)

A three-part series on implementing a tiny Unix-like shell in C, covering parsing, forking processes, exec, pipes, and basic error handling.

• Part 1 — Core loop & parsing
• Part 2 — Processes & exec
• Part 3 — Memory & polish

Tackling LeetCode Bash Problems – A Practical Guide

Summary of how I completed the LeetCode Bash problem set: patterns, useful Unix tools, and thinking in data streams. Includes example one-liners and explanations.

Daily LeetCode Progress — Algorithm Practice Series

A running series logging my daily LeetCode progress (Day 1, 2, 3, …), focusing on patterns, mistakes, and what I learned from each batch of problems in Python and C++.