This is my space where I write about the things I am working on and share what I've learned. My blogs are usually about the experiments I do, and sometimes about my personal experiences. Feel free to explore and reach out if you want to connect!
A month into using Obsidian daily as a second brain, daily notes, a predefined tag system, Kanban, finances and todos all linked together, plus why paper, Notion and Logseq didn't stick, and my plan to build an AI assistant that reads it all back to me.
I built Calmly, a local Python tool that tracks the news sites I trust, detects new posts by diffing homepages, and uses Gemini to filter by topic and rewrite everything into simple English, so I can keep up without feeling dismay.
I built hypr-breaktimer, a Go-based terminal UI break timer for Hyprland, because my previous GUI timer didn’t fullscreen reliably, was easy to bypass across workspaces, and used too much RAM.
Besides working on software and keeping up-to-date with the latest in the technological side, I enjoy hiking, photography(you can find them in this site) and reading sci-fi and thriller books (if you are looking for recommendations, read Blake Crouch’s, Dark Matter, Recursion, and Upgrade, or a Dan Brown). I am also intereted in automobiles, feel free to reach out if you wanna talk anything about cars, motorcycles, engines, and more!
Opinionated Arch + Hyprland setup that turns a fresh install into a fully-tiled, themed dev environment in minutes, my daily driver.
My note-taking, mental map, and knowledge base of operations. Local-first markdown, graph view, plugins for everything, no lock-in.
Zero-shot NER model that pulls any entity type you name without fine-tuning. Lightweight enough to run locally, my go-to for extraction over unstructured text.
Probabilistic record linkage and entity resolution at scale. Dedupes and links messy records across sources, the backbone when identities don't match cleanly.
Building web data extraction pipelines and OSINT tools for generating actionable insights from unstructured data.
St. Xavier's College, Kathmandu, Nepal.
Completed my research thesis on 'Using Machine Learning to Predict Crop Yield in Nepal', from data collection to model training and deployment, under a supervisor
Presented my abstract on Machine Learning in Agriculture at the International Conference held in St. Xavier's College.
Sabda is a mobile app for learning Nepali vocabulary using structured lessons, quizzes, and spaced repetition. Built with Flutter and Firebase, it offers a modern UI, offline support, and secure data handling.
RelevanCite is an offline, privacy-preserving AI assistant that automates citation validation for research using NLP, NLI, and modular architecture. All processing is local, ensuring data security and reproducibility.
A Python-based project that automates the evaluation of OMR answer sheets using OpenCV for image processing and Tesseract for OCR, exporting results in CSV format.
PAYO is a secure, offline-first digital payment solution using one-time QR codes, built with React Native, TypeScript, ExpressJS, and MongoDB. It enables real-time payments with minimal connectivity and robust security features.
You can reach me at aayushshrestha2003@gmail.com