AAheadMint42m ago · 3 min readWhich Self-Supervised Objective Should You Use? A 5-Question Decision Guide"Let's do self-supervised pretraining" is easy to say. Picking which SSL objective fits your data is where projects go right or wrong. Here's the decision process we use, written as five questions. Q1aheadmint.hashnode.dev
AAheadMint21h ago · 3 min readFollowing One Sentence Through a Transformer: A Tensor-Shape WalkthroughDiagrams of the Transformer show boxes and arrows. What helped me most was following a single sentence through the model and writing down the shape of the data at every step. Here is that trace, usingaheadmint.hashnode.dev
OOmar Essam El Sherbiny2d ago · 4 min readAn image classifier you can trust 80% of the time: confidence gating and Grad-CAMThis was my final project for IBM's Deep Learning and Reinforcement Learning course. Fashion-MNIST is a familiar benchmark, so I framed it as a business task instead of a leaderboard entry: an online omaressam.hashnode.dev
OOmar Essam El Sherbiny2d ago · 5 min readTeaching a CNN to read hieroglyphs: 96.9% on 59 signs, and why my first model failedEvery Egyptian hieroglyph has a catalogue number. Egyptologists use Alan Gardiner's sign list, so the owl is G17, the reed leaf is M17, the water ripple is N35. I wanted to see how far a small convoluomaressam.hashnode.dev
PPriya Raman3d ago · 6 min readWhen a Floor Plan Starts Moving, Scale Is a Camera ProblemThere is a quiet shift happening on listing pages and renovation decks: the floor plan no longer sits still. A short clip pans, reveals, or "flies" across the same drawing you used to annotate with a archybase.hashnode.dev
KKarnsinh Jadhav4d ago · 6 min readBuilding PotholeNet: A Computer Vision Project for Road Damage DetectionWhat started as a computer vision project is becoming an exploration of road safety, geospatial intelligence, and real-world software engineering. Road potholes are more than just an inconvenience. Thkaranjadhav.hashnode.dev
AAI Engineering Illustrated4d ago · 17 min readHow Mixture of Experts WorksDeepSeek-V3 stores 671 billion parameters but runs only 37 billion for each token. Here is the router, the balancing tricks, and the GPU traffic that make that possible. Open the model card for almostaiengillustrated.hashnode.dev
SShivam Kumar6d ago · 5 min readInside Kimi K3.1 Pro: How Joint RL Across Hundreds of Domains Built a Near-Frontier Open ModelInside Kimi K3.1 Pro: How Joint RL Across Hundreds of Domains Built a Near-Frontier Open Model Shivam Kumar · October 26, 2026 Moonshot AI's Kimi K3.1 Pro, announced October 23, 2026, is a 3-trillion-shivaysinghrajput.hashnode.dev
HHimanshu MishraOct 24 · 3 min readWhat is Artificial Neural Network (Perceptron Model) ?To understand the topic, let us first understand what is Artificial Neuron. In school days we all have read about biological Neuron. It is a fundamental building block of our nervous system. This bioltechmojo.hashnode.dev
SShivam KumarOct 25 · 5 min readInside Voxtral 3's One-Pipeline Voice Agent: Listen, Think, Act, Speak in Under 500msMost voice agents are three products pretending to be one. A transcription service hears the caller. A language model decides what to say. A speech service says it. Between each hop, latency accumulatshivaysinghrajput.hashnode.dev
