Journey of Curiosity Arshad Kazi

India

Elsewhere

Technical
Blogs

Things I had to understand from the ground up, written down while learning them. Notebooks follow a subject day by day; essays make one argument.

Why I write these

I write these to check whether I actually understand something. If I can explain it from the start, with the reasoning and not just the result, I do. If I can't, the gap shows up on the page.

I'm mostly self-taught, so I don't always know what I'm missing. Writing is how I find out.

I move between subjects because that's how my curiosity works, and because an idea often makes more sense once I've seen it somewhere else.

These are working notes, not tutorials. Some of it will be wrong. Where I'm not sure, I say so.

Vision

Recovering structure from projections. An image discards depth, so every estimate of shape, motion and identity is an inference under geometric and statistical constraints.

Fig. 1.1 Notebook · 3 days · 17.4k words · ongoing

Structure from Motion

Geometry-based 3D reconstruction from images: camera models, epipolar geometry, triangulation and structure from motion.

  1. 01SIFT: detecting and matching keypoints across images19 min
  2. 02Epipolar geometry: why a match has to sit on a line22 min
  3. 03From matches to cameras to points: estimating F, taking E apart, triangulating37 min

Fig. 1.2 Essay · Feb 2026 · 5 min

Temporal and Spatial Attention for 3D Pose Estimation

How temporal and spatial attention mechanisms enable accurate 2D-to-3D human pose estimation, with a deep dive into DSTformer.

  1. 01The Ill-Posed Problem: Lifting 2D to 3D
  2. 02Temporal Attention: Understanding Motion Over Time
  3. 03Spatial Attention: Maintaining Anatomical Consistency
  4. 04Fusion of Temporal and Spatial Streams

Fig. 1.3 Essay · Mar 2024 · 3 min

Building a Sign Language Recognition System with CNNs and OpenCV

A practical walkthrough of building a real-time ASL alphabet recogniser. Dataset, CNN architecture with the actual math, and live webcam inference using OpenCV.

  1. 01Dataset
  2. 02Build and Train the Model
  3. 03OpenCV

Neuroscience

Computation in biological hardware: spiking neurons, hierarchies of receptive fields and dense feedback, doing inference on roughly twenty watts.

Fig. 2.1 Notebook · 7 days · 9.8k words

Understanding Emotion

What are emotions, really? The neuroscience of feeling, and why an autistic brain might process them differently.

  1. 01What even is an emotion?3 min
  2. 02Are guilt and love just layered explanations?3 min
  3. 03Why did evolution bother building emotions?5 min
  4. 04Guilt, cheating, and why monogamy won10 min
  5. 05What is happiness and where does emotional pressure go?7 min
  6. 06Beyond language: the other tools that shape how we feel6 min
  7. 07How emotions control what you think, who you trust, and what you remember11 min

Fig. 2.2 Notebook · 7 days · 5.8k words

Understanding the Visual Cortex

How does the brain process what the eye sees? Tracing the path from retina through two parallel streams, the mechanisms at each stage, and the computations behind perception.

  1. 01How does the brain begin processing vision?2 min
  2. 02What mechanisms power each visual stream?2 min
  3. 03How do neurons detect and assemble edges?2 min
  4. 04How do neurons use context from their neighbours?2 min
  5. 05How does the brain perceive depth?6 min
  6. 06How does seeing become knowing?5 min
  7. 07The complete architecture — how vision flows through the brain9 min

Fig. 2.3 Essay · Apr 2026 · 14 min

The Visual Cortex as a Computer Vision Architecture: Dorsal and Ventral Streams

How the brain's visual cortex uses two parallel streams, feedback loops, and relative perception to process vision. Everything rewritten for engineers who build vision systems.

  1. 01The Split You Did Not Know Was in Your Design
  2. 02The Dorsal Stream: Real-Time Spatial Inference
  3. 03The Ventral Stream: Hierarchical Object Recognition
  4. 04Implications for Computer Vision Engineering

Machine learning

Function approximation at scale. Parameterised models fitted by gradient descent, where the architecture decides what can be represented and the data decides what is learned.

Fig. 3.1 Notebook · 7 days · 3.2k words

Overview of Modern Nets

Revisiting transformers, tokenizers, attention, and the GPT family. Intuitive understanding for interviews and curiosity.

  1. 01The encoder block, BPE, and how tokens are made2 min
  2. 02Multi-head attention and why we use dot products2 min
  3. 03Masking, padding, and BERT vs GPT2 min
  4. 04The transformer architecture, drawn from scratch2 min
  5. 05Pretraining, feedforward, residual connections, and layer norm3 min
  6. 06LLaMA, Mistral, and the road to reasoning models2 min
  7. 07RAG, LangChain, LangGraph, and the LLM tooling landscape5 min

Fig. 3.3 Essay · Jan 2024 · 6 min

Vector Database Management Systems: Architecture and Internals

A technical overview of how Vector Database Management Systems work, covering vectorization, indexing, hardware tradeoffs, and the core challenges in building systems that search by meaning rather than exact match.

  1. 01What is a Vector Database?
  2. 02Why Not a Regular Database?
  3. 03General Architecture of a VDBMS
  4. 04Use Cases
  5. 05Challenges and Limitations

Foundations

Probability, estimation and evaluation: the formal machinery for reasoning under uncertainty, and for deciding whether a model's output can be trusted.

Fig. 4.1 Essay · Feb 2024 · 3 min

ROC Curves and AUC: Picking the Right Threshold for Your Model

Why accuracy fails on skewed data, how confusion matrices work, and how to use ROC curves and AUC to evaluate and compare probabilistic classifiers.

  1. 01Why Accuracy Is Not Enough
  2. 02True Positive, True Negative, False Positive, False Negative
  3. 03Confusion Matrix
  4. 04ROC Curve
  5. 05AUC (Area Under the Curve)

Fig. 4.2 Essay · Nov 2023 · 4 min

Probability Mass Functions, Expectation, and Variance

Relearning probability from a machine learning perspective. What PMFs are, how to compute expectation and variance from first principles, and why any of this matters for ML.

  1. 01What is Probability?
  2. 02Probability Distribution
  3. 03Probability Mass Function
  4. 04Expected Value (Expectation)
  5. 05Variance
  6. 06Why does any of this matter for ML?