New Projects — GRIT, Sage, Terrain Analyser, Self-Healing RAG, and More

Six new projects added to the showcase: a Git reimplementation in Rust, an ML-powered scraper, a lunar terrain analyser, a self-healing RAG system, Google ADK agent exploration, and more.

📋 What’s New

I’ve added six new projects to the portfolio and reordered the showcase by newest commit so the most active repos appear first.

🦀 GRIT — Git Reimplemented In Rust

A from-scratch Git implementation in Rust, byte by byte, with only two dependencies. Covers the core Git object model — blobs, trees, commits, refs — plus init, add, commit, log, diff, branch, checkout, merge, and clone over HTTP.

$ grit init
$ grit add .
$ grit commit -m "first commit"
$ grit log

The goal: understand Git from the inside out by building every command from scratch, with a focus on correctness and performance.

➡️ GitHub

🧪 Sage — ML-Powered E-Commerce Scraper

A sandbox e-commerce scraper that uses ML for price prediction, orchestrated with Prefect workflows and tracked with MLflow experiment tracking. Serves predictions via FastAPI.

scrape → feature engineering → ML price pred → FastAPI serve

Full pipeline observability with Prefect for orchestration, MLflow for experiment tracking, and a clean FastAPI serving layer.

➡️ GitHub

🌙 Terrain Analyser — Lunar Crater Detection

An automated crater and boulder detection pipeline for high-resolution Chandrayaan-2 lunar surface imagery. Uses YOLOv8 with tile-based processing to handle large satellite images.

Key features:

• Tile-based sliding window inference for high-res imagery
• YOLOv8 object detection for craters and boulders
• Geospatial coordinate mapping back to lunar reference frames
• Visualization with detected objects overlaid on original imagery

➡️ GitHub

🔄 Self-Healing RAG

A RAG system that detects and fixes retrieval/generation failures before serving answers. Monitors confidence scores, detects low-quality retrievals, and triggers automatic retry with reformulated queries — so users always get a coherent response.

➡️ GitHub

🤖 MultiAgent ADK — Google ADK Agent Exploration

Exploring Google’s Agent Development Kit (ADK) for building multi-agent AI systems. Investigates agent orchestration patterns, tool delegation, and structured handoffs between specialized agents.

➡️ GitHub

🧠 Persistent Memory — MCP Long-Term Memory Server

An MCP server that gives AI agents persistent long-term memory. Stores facts as 384-dim vector embeddings (all-MiniLM-L6-v2) in SQLite, with retrieval via cosine similarity, keyword search via FTS5/BM25, and tag-based filtering — zero external infrastructure.

➡️ GitHub · Post

🗂️ Showcase Reorganisation

The project showcase is now sorted by newest commit date, so the most actively developed projects appear first. The Gallery section on the landing page and the full projects page both reflect this ordering — the projects page defaults to sorting by last updated.

The full lineup:

Check out the full list on the projects page or browse the GitHub profile.