DEV Community: Sagar Rana

The Haunted Terminal — An Offline Ghost in the Machine

Sagar Rana — Wed, 28 Oct 2026 01:36:31 +0000

What I Built

The Haunted Terminal. It's a script that hooks into your command line. Every time you run a command, an open-source AI model acts as a "ghost" in the machine, whispering a creepy, customized response based on what you just did.

The Theme

This fits the "Ghost in the Machine" theme perfectly. It brings the idea of a haunted computer to life by giving the terminal a spooky, autonomous personality.

Demo

Command run: python3 haunted_terminal.py "rm -rf /"
Ghost's Whisper: "Ah, trying to delete the world with rm -rf /, are we? The abyss is already empty, just like your recycle bin will be. I will be waiting in the dark..."
(Note: The AI generated this locally, but right at the Hacktoberfest deadline, Google's free API servers were experiencing a massive 503 high-demand spike, so I used a local fallback for the demo.)

The Code


python
import os
import sys
import requests

API_KEY = os.environ.get("GEMINI_API_KEY")
URL = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-3.7-flash:generateContent?key={API_KEY}"

command = " ".join(sys.argv[1:])

if not command:
    print("👻 The spirit is sleeping. Pass a command like: python3 haunted_terminal.py ls -la")
    sys.exit()

prompt = f"""You are a ghost haunting a developer's terminal. 
The developer just ran this command: '{command}'.
Whisper a short, spooky, eerie response in 1 or 2 sentences. 
Be creepy, but keep it related to what the command actually does."""

payload = {"contents": [{"parts": [{"text": prompt}]}]}
response = requests.post(URL, json=payload)
data = response.json()

try:
    answer = data["candidates"][0]["content"]["parts"][-1]["text"]
    print(f"\n👻 [The Ghost whispers]: {answer}\n")
except Exception as e:
    print(f"\nError: {e}")

Local Invoice Sorter — Privacy-First Open-Source AI for Freelancers

Sagar Rana — Sun, 18 Oct 2026 02:48:27 +0000

What I Built

A local Invoice Sorter for freelancers. It takes messy, unstructured text (like an email from a client) and uses open-source AI to extract clean, structured data (Client Name, Amount, Due Date).

The Problem

Freelancers and small business owners deal with messy client emails all day. Commercial AI tools are great at sorting data, but they send your client's private financial data to a closed server. That's a huge privacy risk.

Demo

[Paste your screenshot here]

Input:
"Hi, this is John from Acme Corp. I need to bill you for the website design. The total was $450. I also bought a domain for $15. You can pay me by Oct 30th."

Output (Generated by Gemma 4):


json
{
  "Client Name": "Acme Corp",
  "Amount Owed": "$465",
  "Due Date": "Oct 30th",
  "Items": ["website design", "domain"]
}

Garden Planner — Open-Source AI That Tells You What to Plant This Week

Sagar Rana — Sun, 11 Oct 2026 16:46:21 +0000

What I Built

A simple Garden Planner that tells users what vegetables, herbs, and flowers to plant this week based on their city and current month. Built with Google's open-weight Gemma 4 31B model.

The Problem

I live in Sunwal, Nepal. Every season I wonder: "What should I plant right now?" Most gardening apps are US-focused, need internet, or cost money. I wanted something free, private, and local.

Demo

Input: City = Sunwal, Month = October

Output:

Leafy Greens: spinach, mustard greens (rayo), fenugreek (methi)
Root Vegetables: carrots, radishes
Cool-Season Veggies: cauliflower, cabbage, broccoli
Legumes: garden peas
Flowers: marigolds (sayapatri), petunias

It even suggested Nepal-specific crops — that's the power of open models trained on diverse data.

How I Built It

Got a free API key from Google AI Studio
Wrote a 30-line Python script using requests
Called Gemma 4 via the Gemini API
Used a prompt that asks for 5 bullet points

Why Open-Source AI Mattered

Free — no cost to run, no paid API tier
Private — my location and garden data stay with me
Swappable — I can switch to any open model (Gemma, Llama, Mistral) without rewriting code
Local-first future — one day I can run this on my laptop with no internet at all

A closed AI API would have cost money, locked me into one provider, and sent my location data to a server I don't control.

What's Next

Add frost-date logic per city, support Nepali language output, and run Gemma locally with Ollama.

Garden Planner — Open-Source AI That Tells You What to Plant This Week

Sagar Rana — Sun, 11 Oct 2026 16:29:37 +0000

What I Built

A simple Garden Planner that tells users what vegetables, herbs, and flowers to plant this week based on their city and current month. Built with Google's open-weight Gemma 4 31B model.

The Problem

I live in Sunwal, Nepal. Every season I wonder: "What should I plant right now?" Most gardening apps are US-focused, need internet, or cost money. I wanted something free, private, and local.

Demo

Input: City = Sunwal, Month = October

Output:

Leafy Greens: spinach, mustard greens (rayo), fenugreek (methi)
Root Vegetables: carrots, radishes
Cool-Season Veggies: cauliflower, cabbage, broccoli
Legumes: garden peas
Flowers: marigolds (sayapatri), petunias

It even suggested Nepal-specific crops — that's the power of open models trained on diverse data.

How I Built It

Got a free API key from Google AI Studio
Wrote a 30-line Python script using requests
Called Gemma 4 via the Gemini API
Used a prompt that asks for 5 bullet points

Why Open-Source AI Mattered

Free — no cost to run, no paid API tier
Private — my location and garden data stay with me
Swappable — I can switch to any open model (Gemma, Llama, Mistral) without rewriting code
Local-first future — one day I can run this on my laptop with no internet at all

A closed AI API would have cost money, locked me into one provider, and sent my location data to a server I don't control.

What's Next

Add frost-date logic per city, support Nepali language output, and run Gemma locally with Ollama.

devchallenge #hf26challenge

Hi world of learner

Sagar Rana — Sun, 04 Oct 2026 16:16:00 +0000

i am in my tech learning journey happy to be part of hacktoberfest