Mishan Shah
Four years in the industry. The past year and a half as Head of Engineering & IT at LQ Digital, before that leading their development team. I've always been the type who can't leave a new tool or idea alone until I understand it, and that habit is the reason I've ended up working across such a wide range of frameworks in both backend and AI.
I built my own LSTM model for the Nepal Stock Exchange, and later an AI-driven meeting intelligence tool. Both are open source on GitHub, @mishansavy.
The world keeps changing, and if you stand still you fall behind far enough that catching up gets hard. That's why I never stop learning or sharpening what I know. AI can write boilerplate for you and dig through thousands of lines to find a bug, but knowing what to accept from it and what to throw out, that's the real game.
Kathmandu, NepalMeetMind
↗Meeting intelligence for teams: WebRTC video calls with live captions, Whisper transcription, and automatic action-item extraction with spaCy.
NEPSE Trend Analysis
↗An LSTM model trained on Nepal Stock Exchange data. Mostly an exercise in seeing where the model falls apart on a thin, illiquid market.
JWE Decryption Demo
↗A small demo where JWE payloads get decrypted in the browser. The private key never leaves the client, so the server only ever sees ciphertext.
Employee Management
↗A few FastAPI services that talk to each other over signed JWTs, wired together with Docker. Built it to get the service-to-service auth right.
Why do humans learn faster than LLMs? Exploring core knowledge, embodied learning, and world models and why AI architecture, not scale, may be the real limit.
Why I Still Use NGINX Over Everything Else
Jul 11, 2026One NGINX config written by hand handled 2,500 requests per second on a single 2vCPU/8GB VPS with zero errors. Here is why I still choose NGINX over Caddy and Traefik.
Curious about systems and how things actually work underneath.
Happy to talk backend, infra, or research direction. Drop a message or reach out directly.