Inspiration
Engineers waste weeks flipping through spreadsheets and datasheets, often defaulting to familiar materials out of habit. We built MaterialMind to be an intelligent co-pilot—combining verified engineering data with generative AI reasoning to slash material selection from weeks to minutes.
What it does
MaterialMind is an end-to-end AI decision engine. Users describe their application via a 7-step wizard (temperature, environment, strength, weight, budget, quantity). The system cross-references a verified material database (Pandas/CSV), uses Google Gemini to analyze trade-offs, and recommends the optimal material. It also compares multiple materials side-by-side, gives specific manufacturing advice, and exports professional PDF comparison reports.
How we built it
Backend: Python FastAPI serving REST endpoints, with Pandas for data ingestion and ReportLab for PDF generation.
AI Layer: Google Gemini API for comparative reasoning, manufacturing guidance, and fallback recommendations.
Frontend: React with Vite, styled with Tailwind CSS (blueprint/engineering aesthetic).
Integration: Seamless REST communication between the React wizard and FastAPI, with real-time API health polling.
Challenges we ran into
Prompt engineering: Forcing Gemini to provide balanced trade-offs and weaknesses, rather than declaring a single unqualified "winner."
Data wrangling: Cleaning inconsistent numeric fields (density, tensile strength) from the raw engineering CSV.
CORS & local dev: Coordinating the Vite dev server (5173) with the FastAPI backend (8080) without issues.
PDF layout: Getting ReportLab to gracefully handle multi-page AI-generated text alongside structured property tables.
Accomplishments that we're proud of
A fully functional MVP covering the entire workflow: Search → Recommend → Compare → Manufacture → Report.
A distinct, technical UI/UX (blueprint grid, corner-bracket spec cards) that resonates with engineers.
Robust fallback logic—the system still provides data-backed recommendations even if the Gemini API is unavailable.
Real-time API status indicator in the header, giving users immediate feedback on system health.
What we learned
How to effectively pair structured tabular data (CSV) with unstructured LLM reasoning to produce actionable insights.
The importance of iterative prompt refinement to get honest, comparative engineering advice rather than generic AI fluff.
How to build a clean, maintainable monorepo-style project with decoupled frontend/backend communicating via strongly typed Pydantic schemas.
What's next for MaterialMind AI
Expand the database: Scale from 72 verified materials to 1,000+ (polymers, composites, ceramics).
Image-to-material: Allow users to upload a photo of a broken part and get a material match.
CAD integration: Plug directly into Onshape/Fusion 360 to suggest materials based on 3D geometry stress analysis.
Lifecycle scoring: Add sustainability metrics (carbon footprint, recyclability) to every recommendation.
Log in or sign up for Devpost to join the conversation.