“The Last Commit: an offline AI séance for abandoned software.”

aditi — Sat, 31 Oct 2026 09:06:54 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

The Last Commit is an AI-powered repository history explorer.

Every repository carries the ghosts of its past: abandoned ideas, first experiments, renamed files, and the quiet initial commit that started everything. Usually, those stories stay buried in commit logs and README files.

I built The Last Commit to bring those ghosts back to life.

You give it a GitHub repository, and it:

Clones the repository
Extracts evidence from its README and commit history
Answers your questions using a locally running Gemma model
Provides evidence-backed citations such as [E1] and [E2]
Refuses to invent facts that are not present in the repository evidence

For example, if you ask, “What commits exist in this repository?”, it answers using only the actual commit evidence and labels each claim with its source.

I interpreted the Ghost in the Machine theme as giving a voice to the forgotten history inside a codebase. The “ghost” is not a person—it is the repository’s past.

Demo

Local demo:

http://127.0.0.1:8000

The app includes a simple web interface where you can enter a question about a repository and receive an evidence-based answer.

Example:

{
  "repo": "juice-repo",
  "question": "What commits exist in this repository?"
}

Response:

The repository contains the following commits:
- Add files via upload [E1]
- Initial commit [E2]

Code

aditithakur01 / the-last-commit

AI-powered repository history explorer with evidence-based answers using Gemma and FastAPI.

🕯️ The Last Commit

The Last Commit is an AI-powered repository history explorer.
It clones a GitHub repository, extracts factual evidence from its README and commit history, and answers questions using a locally running Gemma model through Ollama.

✨ Features

Clone any public GitHub repository
Extract repository evidence from README and commits
Answer user questions using local Gemma via Ollama
Provide evidence-backed answers with citations such as [E1]
Avoid inventing facts not present in the repository evidence
Simple web interface built with HTML and FastAPI

🧠 How It Works

User provides a repository name.
The app clones the repository.
It creates an evidence index from the README and commit messages.
The user asks a question through the web UI.
Gemma answers using only the available evidence.
The response includes evidence labels such as [E1] or [E2].

🛠️ Tech Stack

Python
FastAPI
Ollama
Gemma LLM
HTML, CSS, JavaScript
Git

🚀

…

View on GitHub

How I Built It

The project is built around open-source AI and local inference.

Gemma runs locally through Ollama as the reasoning model
FastAPI powers the backend and exposes the /ask endpoint
Git is used to clone repositories and read commit history
A lightweight evidence index is generated from README content and commit messages
The frontend is built with plain HTML, CSS, and JavaScript

The key design decision is that Gemma does not answer freely. Instead, it receives repository evidence and is instructed to answer only from that evidence. Every factual claim must include an evidence label.

This makes the system more trustworthy: it separates facts from interpretation and reduces hallucination.

Why Does Open Innovation Matter?

Open innovation is the foundation of this project.

Because Gemma runs locally through Ollama, The Last Commit can work without sending repository contents to a closed cloud API. The repository data stays on the user’s machine.

It also means:

No per-request API cost
No dependency on a single closed provider
The model can be swapped for another open-weight model later
The project can run offline once the model is downloaded
Developers can inspect, modify, and improve the entire pipeline

For a tool that reads private or personal repositories, keeping inference local is not just a technical choice—it is a privacy choice.

My Agent Session

I built this project iteratively with an AI-assisted development workflow: setting up repository cloning, generating the evidence index, connecting Gemma through Ollama, building the FastAPI endpoint, and creating the web UI.

The most important moment was when the model correctly responded using only the provided evidence and cited [E1] and [E2] instead of inventing details.

The Spirit of It

Every repository begins with an initial commit.

Most people never think about it again. But that first commit is where the project’s story begins—the first file uploaded, the first idea made real, the first trace left behind.

The Last Commit treats that history with respect. It does not invent a dramatic backstory for a repository; it simply listens to the evidence that remains and helps you understand the project’s past.

Prize Categories

Best Use of Gemma

LazyChat Summarizer: A Local Open-Source AI Agent for Effortless Group Chat Summaries

aditi — Fri, 23 Oct 2026 15:30:12 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

DEV Community: aditi

“The Last Commit: an offline AI séance for abandoned software.”

What I Built

Demo

Code

aditithakur01 / the-last-commit

AI-powered repository history explorer with evidence-based answers using Gemma and FastAPI.

🕯️ The Last Commit

✨ Features

🧠 How It Works

🛠️ Tech Stack

🚀

How I Built It

Why Does Open Innovation Matter?

My Agent Session

The Spirit of It

Prize Categories

LazyChat Summarizer: A Local Open-Source AI Agent for Effortless Group Chat Summaries

What I Built

Demo

Code

How I Built It

Why Does Open Innovation Matter?

My Agent Session

The Lazy Payoff

Prize Categories