Markov Chain Lab
Interactive learning platform for Markov chains, automata, and grammar conversion.
leptons1618@anishgiri.dev:~/projects ls --sort=year --all
The work — production systems, experiments, and everything in between, filterable by category, tag, and stack.
Interactive learning platform for Markov chains, automata, and grammar conversion.
A modular OCR pipeline that turns restaurant menu photos into schema-compliant JSON — sections, groups, items, and prices — with column-aware layout detection and global price-item matching.
Paste a YouTube link, get a transcript you can search and click to hear — Whisper already returns timestamps, so every line becomes a seek target back into the video.
Personal portfolio site for projects, case studies, and journal posts with a browser-based admin surface and AI authoring assistant. Content lives in Cloudflare D1, so publishing is a save rather than a commit.
A version-controlled AI engineering environment: agent skills pinned to exact commits, install profiles, and bootstrap scripts that leave upstream repositories untouched.
Daystorm fuses camera, CAN bus, radar, and cabin audio onto a causal 2 Hz grid so a language model answers grounded questions about a two-second driving window.
A browser-based CAD application with a drawing engine, authenticated user management, and DXF/DWG interop. Built for engineers who need CAD workflows in the browser.
A Microsoft Edge/Chromium extension that lets you log personal Jira worklogs via the Jira REST API, with quick logging, issue search, reminders, and optional automatic logging.
A Go app for generating, managing and testing SSH keys, with a desktop GUI and a terminal UI over the same core — it shells out to OpenSSH, and the private key never leaves the machine.
A browser extension that shows Bengali word meanings, transliteration, and part of speech in a popup when you select text on any website. Backed by a Fastify API with a ~15k-entry dictionary.
An Ignition module that docks an AI assistant into the Designer. The Designer holds a chat panel and nothing else — every model call crosses an RPC boundary to the gateway, and nothing reaches a project without a person approving it.
Proofline turns an audit or investigation into one self-contained HTML file — three visual modes, a print-safe layout, and an embedded JSON block that agents can parse and that never drifts from the prose.
A local-first system monitoring agent that tracks CPU and memory, stores events in SQLite, exposes a JSON-RPC CLI over Unix sockets, and optionally enriches insights via a local Ollama model — all within a 200 MB memory budget and a systemd service.
Local-first desktop monitor for a face detection agent — Tauri shell serves a Dioxus WebAssembly UI while a Python pipeline runs YuNet ONNX detection and online clustering on Jetson, syncing via TCP JSONL IPC.
A browser workspace that turns Excel workbooks into database tables — and then lets you clean, validate and reshape the data before any of it is committed.
Chat with your own documents without anything leaving the machine — Streamlit in front, LangChain and ChromaDB behind, and a local Ollama model doing the thinking.
Evaluation harness for ML experiments with standardized metrics and reports.
Conversational AI frontend with multi-model support and conversation management.
Face detection and recognition for live camera feeds — one build that runs on a Jetson, a CUDA desktop or a laptop CPU, with roles, audit and multi-camera sources.
Point your camera at a Sudoku puzzle and it solves it on the spot — a TensorFlow.js CNN that reads the grid in the browser, then ships as an Android app and a PWA.
A browser tool for turning an SVG drawing into an AutoCAD-style DXF block — click a base point on a canvas, set the attributes, and get real DXF entities out the other end.
A local-first chat client for Ollama — switch between five open models mid-conversation, talk to it or type to it, and keep every token on the machine it runs on.
Cross-platform mobile watch-party app with realtime room sync, queue, and chat.
Synchronised watch parties with no account and nothing to install — an Expo client sharing playback over WebRTC data channels, with one small WebSocket server holding the room together.
Explainable and trust-aware image classification prototype with uncertainty and OOD analysis.
Computer vision pipeline for estimating vehicle speed from traffic video.
Browser-based computer vision playground using TensorFlow.js for in-browser inference.
A Discord bot that plays Akinator — you think of a character, it asks twenty questions and guesses. Five Python files and a Heroku worker.
Ask about Anish
Answers come from this site's own pages.
Past conversations
Kept only in this browser — clearing site data forgets them.
This is a language model reading published pages. It can be wrong — the pages themselves are the record.