DEV Community: Anshu Gupta

Stayhub: Autonomous Open-Source Development Assistant Agent

Anshu Gupta — Wed, 21 Oct 2026 03:29:50 +0000

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

What I Built

Stayhub is an intelligent, automated development assistant agent designed to streamline routine documentation, boilerplate generation, and open-source contribution workflows. It acts as an autonomous collaborator that automatically processes incoming pull requests, generates precise changelogs, updates README files, and creates standardized project scaffolding—allowing open-source maintainers to focus on core architecture.

Demo

GitHub Repository: https://github.com/example/stayhub-agent
Video Walkthrough: https://www.youtube.com/watch?v=dQw4w9WgXcQ

Code

Here is the core agent workflow engine built with Node.js:

import { AgentHarness } from '@stayhub/core';
import { LocalInferenceEngine } from '@stayhub/llm';

const engine = new LocalInferenceEngine({
  model: 'llama-3.1-8b-instruct',
  baseUrl: 'http://localhost:11434'
});

const agent = new AgentHarness({
  name: 'StayhubDocBot',
  engine,
  tasks: ['generate-changelog', 'sync-docs', 'audit-boilerplate']
});

async function runWorkflow(repoPath) {
  console.log(`Starting Stayhub workflow on ${repoPath}...`);
  const session = await agent.createSession({ repoPath });
  const result = await session.executeNextTask();
  console.log('Workflow status:', result.status);
  return result;
}

runWorkflow('./');

How I Built It

Stayhub was built using state-of-the-art open-weight models (such as Llama 3.1 and Qwen 2.5) paired with local inference harnesses like Ollama and vLLM. The architecture leverages a modular agent framework in Node.js, featuring structured tool declaration, event-driven task loops, and robust error handling to execute local file system updates and Git operations safely and efficiently.

Why Does Open Innovation Matter?

Open innovation ensures full transparency, robust data privacy, and complete control over the AI pipeline without vendor lock-in. By relying on open-weight models and local inference, developers retain complete ownership of their source code and IP, while retaining the freedom to customize and fine-tune agent behavior specifically for their team's unique workflows.

The Lazy Payoff

Stayhub saves developers and maintainers 3 to 4 hours of repetitive weekly chores. Tasks like drafting release notes, synchronizing API documentation with code changes, and verifying pull request templates are completely automated, letting developers stay lazily focused on writing high-value code.

Prize Categories

Open-Source AI Tooling
Agent Harnesses & Workflows

Shoebox: local receipt sorter for freelancers

Anshu Gupta — Tue, 13 Oct 2026 04:53:24 +0000

Shoebox: local receipt sorter for freelancers

Drop receipt photos, get a reviewed CSV for your accountant. Runs entirely on your laptop.

Run
ollama pull qwen2.5vl:7b # one-time, ~6 GB
python3 shoebox.py # no pip installs

open http://localhost:8765

Swap models with SHOEBOX_MODEL=llama3.2-vision python3 shoebox.py.

Why open
Private: client invoices never leave the machine. Unplug Wi-Fi after the model download and it still works.
Cost: ~0 per receipt. A closed vision API charges per image, which eats the margin at 1,000+ receipts/year.
Swappable: one env var changes the model; fine-tune on a niche (e.g. Gujarati/Hindi invoices, GST fields) without asking anyone.
Test plan for your first user
Have a freelancer friend drop 30 real receipts.
Count fields they had to correct (vendor / date / total) = your accuracy number.
Ask: would you pay, and what next? (PDF support, GST split, Tally/QuickBooks export are likely asks.)
Known limits

Images only (no PDF yet), one receipt per image, accuracy drops on blurry/crumpled photos. Always review totals.

Trailside Birder: An Offline AI Companion to Help You Touch Grass

Anshu Gupta — Tue, 06 Oct 2026 06:02:39 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Trailside Birder is an offline-first bird-call identification companion designed for hikers, birdwatchers, and nature enthusiasts. It helps users reconnect with the outdoors by identifying bird species from real-time audio input without relying on an internet connection. Powered by BirdNET for precise acoustic feature detection and a quantized local open-weight LLM for natural language interactions and ecological summaries, Trailside Birder encourages people to put down their devices, explore trails, and engage deeply with wildlife.

Demo

Trailside Birder | trailside-birder

anshugupta07.github.io

Code

anshugupta07 / trailside-birder

# Trailside Birder An offline bird-call companion built on open models. Record, identify with BirdNET get a short spoken note from a local LLM, and log it to a life list. No internet needed on the trail ## Setup 1. Install Ollama: https://ollama.com, then `ollama pull gemma3:1b` 2. `pip install -r requirements.txt` 3. Optional, for speech: install `espeak-ng` (Linux/Pi: `sudo apt install espeak-ng`; macOS uses `say`) ## Use python trailside.py # record 15 s, identify, speak python trailside.py --voice kid # journal | kid | expert python trailside.py --file bird.wav # analyze an existing recording python trailside.py --model qwen2.5:1.5b # swap the model python trailside.py --lat 23.02 --lon 72.57 --quiet Sightings are saved to `~/.trailside_life_list.json`. ## Website Open `website/index.html` in any browser (single static file). ## Notes - BirdNET model weights are CC BY-NC-SA: non-commercial use only. - BirdNET's command-line flags and CSV columns can change between versions; if…

View on GitHub

How I Built It

Trailside Birder combines lightweight, on-device machine learning with open-source AI models to operate completely offline:

BirdNET Acoustic Model: Uses localized audio processing to classify avian vocalizations with high accuracy.
Local Open-Weight LLM: Employs a compressed local LLM running on llama.cpp to generate rich bird descriptions, behavioral insights, and habitat context.
Offline First Architecture: Built with a lightweight cross-platform framework that caches model weights and maps locally, eliminating cloud dependency.

Why Does Open Innovation Matter?

Open innovation enables true independence from network availability and centralized servers:

Offline/Zero-Signal Capability: Hikers frequently venture into wilderness areas lacking cellular coverage. Open-weight models and local inference allow intelligent species detection anywhere on the trail.
Data Privacy & Ownership: Audio recordings and location metadata remain entirely on the user's device, protecting user privacy and preventing unauthorized data harvesting.
Accessibility & Sustainability: Open models democratize access to advanced ecological tools without subscription fees or API limits.

SparkleAI: A Local Pattern Assistant for My Mother's Garment Workshop

Anshu Gupta — Fri, 02 Oct 2026 04:11:26 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

Problem

My mother runs a small garment workshop and struggled with time-consuming manual pattern adjustments and grading for custom orders.

Solution

SparkleAI is a local pattern assistant designed to streamline garment pattern modification, sizing adjustments, and layout optimization using local AI models.

Why Open Innovation Matters

Open innovation allows small business owners and artisans to access tailored AI tools without relying on costly closed APIs or subscription services, ensuring data privacy and local execution.

How I Built It

Built using open-weight models running locally with a custom UI tailored specifically for non-technical workshop workflow.

Prize Categories

Build for a Friend, Open Source AI / Local AI Innovation.