This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built LastFix, a personal troubleshooting memory system for two people close to me, my friend Tilak and my sister Saumya.
They regularly run into small but frustrating technical problems: Wi-Fi disappearing after sleep, printers going offline, external monitors not being detected, Git conflicts, and other issues they have already solved before.
The frustrating part is that they usually remember having fixed it, but not how they fixed it.
So they end up restarting everything, searching Google, asking an AI, and repeating steps that may have already failed.
LastFix solves this differently.
It doesn't try to be another generic troubleshooting chatbot.
It remembers what you already tried.
When a problem happens, LastFix records:
- What went wrong
- The device and context
- What was tried
- What failed
- What worked
Then, when the same problem happens again, you can ask:
"Have I had this problem before?"
LastFix searches your previous incidents and brings back your own experience.
For example:
Previously tried: Restart laptop — Failed
Confirmed fix: Reset network adapter — Resolved the problem
It also provides an Evidence Trail explaining why a previous incident was considered relevant, instead of presenting an unexplained AI answer.
Demo
Watch the complete demo:
Code
GitHub repository:
https://github.com/VaibhavSoni24/LastFix
How I Built It
The core of LastFix is Gemma 3 4B, Google's open-weight model, running locally through Ollama.
The stack is:
- React + Vite
- FastAPI
- SQLite + FTS5
- Gemma 3 4B + Ollama
Gemma is used for three important parts of the application:
1. Memory Extraction
A user can describe a problem naturally:
"My Wi-Fi disappeared after sleep. Restarting didn't help, but resetting the network adapter fixed it."
Gemma converts that into structured incident data and separates successful, failed, and unknown attempts.
2. Query Understanding
A user doesn't need to use technical terminology.
A query such as:
"My screen isn't showing again"
can be expanded into relevant concepts such as monitor, display, HDMI, USB-C, dock, and graphics before searching the user's memories.
3. Memory Reasoning
After retrieving relevant historical incidents, Gemma synthesizes them and presents the previous successful fixes, previous failed attempts, and the evidence behind the result.
The application is local-first, so the troubleshooting history can stay on the user's machine.
Why Does Open Innovation Matter?
Troubleshooting history can contain information people may not want to send to a remote AI service: device details, software environments, network information, configuration details, and personal workflows.
Using an open-weight model locally makes a different architecture possible:
The AI can run where the memory lives.
With Gemma 3 4B through Ollama, LastFix can process and reason over this personal history without requiring a cloud AI API for its core workflow.
Open models also give developers more control over how the system is run, inspected, adapted, and integrated into their own applications.
For LastFix, that isn't just a technical choice. It is part of the product's purpose.
My Agent Session
Optional — not included.
Prize Categories
Overall Winner
LastFix is a qualifying submission for the overall Hacktoberfest Weekend Challenge.
Best Use of Gemma
Gemma is not an optional chatbot layer in LastFix. It powers the core workflow: converting natural-language troubleshooting experiences into structured memories, understanding future queries, and reasoning over the user's retrieved history.
Built during the Hacktoberfest 2026 Weekend Challenge: Build for a Friend.
Top comments (0)