🪨 RIVERLITHOSCOPE: An AI Geological Advisor Built with Gemini

This post is my submission for DEV Education Track: Build Apps with Google AI Studio.

What I Built

I set out to build RiverLithoscope, an interactive web app that acts as an AI geological advisor. It performs source-to-sink analysis on geological images, helping users—from students to exploration geologists—interpret rock specimens, hydrothermal veins, and river contexts.

The Core Prompt Used in Google AI Studio:
I engineered a detailed system instruction to guide the model's analysis.The prompt was structured to make the AI act like a Field Geologist, focusing on:

"You are a senior Field Geologist specializing in Petrology and Economic Geology. Analyze the provided image. First, identify the host rock type and any cross-cutting features (like veins). Describe the mineralogy, texture, and weathering patterns. Then, interpret the geological context: is this a bedrock outcrop, a clast, or part of a river system? Assess its significance for placer deposits or indicator minerals. Structure your output with clear headings for 'Identification Summary,' 'Transport History,' and 'Economic Significance.' Provide a confidence level for your assessment."

Features Utilized:

· Multimodal Reasoning: I used Gemini's ability to analyze visual data (uploaded field photos) combined with text prompts.
· System Instructions: The detailed persona and task list ensured the output was structured and professional.
· Grounding & Heuristics: The app includes a "sensitivity slider" that adjusts the heuristic filter, allowing users to balance between strict, factual analysis and more speculative, exploratory interpretations.
· Frontend Integration: The app, built with React and TypeScript, calls the Gemini API to display the analysis in a clean, card-based interface.

Demo

🔗 GitHub Repository: https://github.com/rajamuhammadyasinkhan2019-lgtm/River-Lithoscope

Here's a walkthrough of using RiverLithoscope in the field:

Step 1: The App Interface
The main dashboard provides access to teaching,professional, and exploration modes. The analysis sensitivity slider (currently set to Balanced at 60% strict) allows you to control how speculative versus conservative the AI's interpretation will be.

Step 2: Capturing Geological Data
I uploaded a field photograph of a dark,fine-grained rock with a prominent reddish-orange vein. The interface allows you to capture both images and field observations before initiating cloud analysis.

Step 3: Cloud Analysis in Progress
After clicking"START CLOUD ANALYSIS," the app begins processing the image through Gemini's multimodal capabilities, combining the visual data with local heuristics and the sensitivity settings.

Step 4: Results with Confidence Score
The analysis completes with remarkable detail!Gemini correctly identified the host rock as "mafic to ultramatic" (likely basalt), recognized the hydrothermal vein with iron-oxide staining, and even provided a Placer Probability Score of +55% (Moderate). The output includes identification summary, transport history, mineral assessment, and economic significance—all with confidence levels.

The final output demonstrates the app's value for real-world exploration: it correctly classified the site as a "Primary Source Zone" and noted that while the outcrop itself isn't a placer, the iron oxides act as "pathfinders" for precious metals downstream—exactly the kind of actionable insight Field Geologists need.

My Experience & Key Takeaways

Working through this track with Google AI Studio was an incredibly rewarding experience. Here are my main takeaways:

1. Prompt Engineering is Everything: The quality of the AI's output depended almost entirely on the clarity of the prompt. I learned that by giving the AI a specific persona ("Senior Field Geologist") and a detailed task list, the analysis went from generic to highly specialized and useful. Refining the prompt to include "confidence levels" was a game-changer for making the tool trustworthy.
2. Multimodal Power is Stunning: The most surprising moment was seeing Gemini correctly interpret a complex geological image—identifying a cross-cutting hydrothermal vein and its iron-oxide staining—and then connecting that observation to broader concepts like "pathfinder minerals" for exploration. It wasn't just describing the picture; it was reasoning about its geological significance.
3. Balancing Speculation with Ethics: Implementing the "analysis sensitivity" slider was a fun challenge. It taught me how to design human-AI interaction where the tool can be both a strict tutor and a creative exploration partner. It also reinforced the importance of the app's Geological Safety & Ethics section—AI is a powerful tool, but it must be used responsibly, respecting local laws and environmental safety.
4. From Concept to Deployed App: Integrating the Gemini API into a React app was smooth. The AI Studio provided a great sandbox to test the prompt, and exporting it to a functional web app made the entire "build" process tangible and fast.

This project has completely shifted how I think about building with AI. It's not just about getting an answer; it's about designing a conversation and a tool that augments human expertise. I'm excited to continue refining RiverLithoscope and explore its potential in educational and field settings.