🌿TraceBack: Step Outside, Capture Something, Trace It Back

#devchallenge #hf26challenge

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Submission 🌿

🌿 TraceBack: Step Outside, Capture Something, Trace It Back

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

What I Built

What if discovering something interesting outside didn't end with taking a photo and forgetting about it?

I built TraceBack — Field Evidence Explorer, an open-source AI-powered project designed to turn real-world observations into evidence-backed discoveries.

Whether you're exploring a nature trail, spotting an unfamiliar bird, discovering a plant, or noticing an interesting landmark, TraceBack helps you investigate what you found and understand the evidence behind it.

Here's how it works:

🌿 Capture: Take a photo or record an observation while exploring the real world.

🧠 Analyze: An open-weight AI model analyzes the observation and suggests a possible identification or explanation.

🔎 Trace: SerpApi searches the web for relevant sources and information related to the observation.

📚 Explore the evidence: TraceBack organizes the results into an evidence trail, making it easier to understand supporting information, source disagreements, and what still needs verification.

🗂️ Save the discovery: Keep observations and their evidence in a personal field journal.

The idea is simple: use AI to make the real world more interesting, not to keep people glued to their screens.

Instead of endlessly scrolling through content, users can step outside, observe something meaningful, capture it, and return to their exploration while TraceBack helps them investigate their discovery.

Demo

🌐 Live Demo: https://traceback-field-evidence.onrender.com

🎥 Demo Video: https://drive.google.com/file/d/1CPMQ-E-DqqNOLaA-SBmad4srLdbtq8Lw/view?usp=drive_link

The demo walks through the journey from capturing a real-world observation to analyzing it with AI and discovering relevant evidence online.

Code

💻 GitHub Repository: TraceBack — Field Evidence Explorer

The project is designed around a modular architecture, separating AI analysis, web search, evidence processing, and field-journal functionality.

How I Built It

I designed TraceBack around two complementary technologies: open-weight AI for observation analysis and SerpApi for online evidence discovery.

🧠 Open-Weight AI

The project is designed to integrate with an open-weight model such as Gemma through Ollama.

The AI analysis layer helps interpret observations, extract useful characteristics, and generate search queries for further investigation.

Keeping the model integration modular makes it possible to experiment with different models rather than locking the entire project to a single proprietary AI provider.

🔎 SerpApi

SerpApi powers the online research component.

It helps retrieve relevant web results for AI-generated queries, allowing TraceBack to collect source links and organize discoverable information around an observation.

Rather than presenting an AI-generated answer as unquestionable truth, the goal is to connect the answer to accessible evidence and make uncertainty visible.

🌱 Local-First Exploration

The architecture also supports saving field observations locally and keeping them available when connectivity is limited. Online evidence searches require an internet connection.

The long-term goal is to make field capture as lightweight as possible, so users can spend more time exploring and less time interacting with the interface.

Tech Stack: React, TypeScript, Tailwind CSS, Express.js, Vite, Featherless API, Ollama (Gemma 3 open-weight AI model), and SerpApi.

Why Does Open Innovation Matter?

For TraceBack, open innovation is not just about using different tools. It's about giving developers more control over how AI-powered exploration works.

Open-weight models make it possible to experiment with local inference, evaluate alternative models, and adapt the analysis pipeline without depending entirely on a closed AI API.

A modular AI architecture also makes the system easier to extend and improve. Different models can be evaluated for different observations, and the project can evolve alongside the open-source ecosystem.

Privacy matters, too. A local-first approach can help keep observations on the user's device during initial processing, while web evidence discovery remains an explicit online step.

SerpApi complements this approach by connecting field observations with information available across the web.

My goal is to combine the flexibility of open AI with searchable evidence, creating a tool that encourages curiosity, supports investigation, and helps people reconnect with the world around them.