This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine.
I am a B.A. History student at Burdwan University. For this challenge, I wanted to build something that connects my interest in History with my interest in AI.
My project is called Itihaser Bhoot (ইতিহাসের ভূত). It is an offline AI assistant that acts like a ghost of a historical figure. It answers History questions using only my own notes.
The idea is simple: an AI that knows what I have studied, shows where its answers come from, and admits when it does not know something.
What I Built
Itihaser Bhoot is a local History question-answering system.
It has three main goals:
- Answer questions using only my personal History notes.
- Quote the source passage so I can check the answer.
- Stay silent or refuse to answer when the notes do not contain enough information.
I wanted to avoid answers that sound convincing but are not supported by my study material.
Demo
The project runs locally on my laptop.
Example workflow:
- I ask a History question.
- The system searches my notes for relevant passages.
- Gemma generates an answer based on the retrieved material.
- The answer includes the source passage.
- If the notes do not cover the question, the system refuses to answer.
The goal is not to make the AI sound knowledgeable about everything. It is to make it reliable within the limits of my notes.
Code
GitHub repository: [Add your repository link here]
The repository contains the code and instructions for running the project locally.
How I Built It
I used Gemma 3 1B through Ollama for local language generation.
For retrieving relevant passages from my notes, I used BM25. It helps find text that matches the question before the model generates an answer.
The system follows a retrieval-based approach:
- My notes are the knowledge source.
- BM25 retrieves relevant passages.
- Gemma 3 1B generates an answer from the retrieved context.
- The source passage is included for verification.
- Questions without sufficient supporting material should be refused.
I built this on a laptop with 3.6 GB of RAM.
I wanted to see whether a small local model could be useful for a focused academic task without relying on a cloud API.
Why Open Innovation Matters
For me, open innovation means making useful technology accessible to people who do not have powerful computers or paid AI subscriptions.
Itihaser Bhoot has a few practical advantages:
- Offline: It runs without an internet connection once the required model is available locally.
- Free per question: There is no cloud API charge for asking questions.
- Runs on my laptop: I do not need expensive hardware for this focused task.
- Private notes: My study notes stay on my own device.
As a student, these things matter to me. I want to experiment with AI while keeping control over my own study material.
What Went Wrong
The project is still experimental, and I faced several problems.
The biggest challenge was answer quality. A small model can sometimes misunderstand a retrieved passage or produce an answer that goes beyond the available evidence.
Bengali also introduced additional difficulties. Bengali questions and notes do not always produce reliable retrieval results. The model can struggle with Bengali phrasing, historical terminology, and generating natural Bengali answers.
Another challenge is refusal behavior. Even when the notes do not contain an answer, the model may try to generate one. I need to improve the system's ability to recognize when the retrieved evidence is insufficient.
I am still testing these limitations rather than claiming they are solved.
Prize Category: Best Use of Gemma
I am submitting Itihaser Bhoot for Best Use of Gemma.
I used Gemma 3 1B as a local language model for a specific academic purpose. The project is not intended to replace textbooks, teachers, or historical research.
It is an experiment in building a small, private, and evidence-based History assistant using open tools.
I hope to improve its retrieval quality, Bengali support, and refusal behavior as I continue working on it.
Hacktoberfest #devchallenge #hf26challenge #opensource #ai
#devchallenge #hf26challenge #software #coding #development #engineering #inclusive #community
Itihaser Bhoot: An Offline Ghost of History
