Future Earth

project: name: Future Earth

inspiration: | We live in an era of "Greenwashing" and decision paralysis. While 73% of global consumers say they would change their consumption habits to reduce their environmental impact, most struggle to bridge the gap between intention and action.

Labels are confusing, supply chains are opaque, and researching a single product takes hours.

We asked ourselves:
"What if we could give every consumer a Ph.D. level sustainability scientist in their pocket?"

We were inspired to build Future Earth to remove the friction from sustainable living.
By moving beyond static barcodes and leveraging the reasoning capabilities of Agentic AI,
we decode the visual world to provide instant transparency and actionable,
planet-positive alternatives.

what_it_does: description: > Future Earth is an intelligent, multimodal eco-scanner that helps consumers understand the environmental impact of everyday products and discover better alternatives.

flow:
  - step: Point & Scan
    details: >
      Users snap a photo or upload an image of a product such as a plastic water bottle,
      polyester shirt, or packaged food item.

  - step: Visual Analysis
    details: >
      The AI identifies the product, probable materials, brand context,
      and packaging type from the image.

  - step: Lifecycle Reasoning
    details: >
      Performs a rapid Life Cycle Assessment (LCA) estimation considering
      extraction, production, transport, and disposal.

  - step: Know Phase
    outputs:
      - eco_score: "1-100"
      - estimated_carbon_footprint: true
      - material_breakdown: true

  - step: Agentic Recommendations
    details: >
      Suggests "Choose Better" alternatives by identifying similar products
      with lower environmental footprints, better packaging, or ethical sourcing.

how_we_built_it: architecture: > A pipeline combining computer vision, vector search, and agentic reasoning.

frontend:
  technologies:
    Gradio 6
  focus: "Responsive, mobile-first experience"

multimodal_vision:
  models:
    - claude-sonnet-4-20250514

  capabilities:
    - texture_analysis
    - label_reading
    - recycling_symbol_detection

orchestrator:
  frameworks:
    - LangChain
    - LlamaIndex
  agentic_workflow:
    - Identify
    - Retrieve_Data
    - Calculate_Score
    - Find_Alternative

rag:
  vector_databases:
    - Pinecone
    - Weaviate
  data_sources:
    - sustainability_reports
    - material_science_datasheets
    - waste_management_guidelines

challenges: - title: Hallucinations vs Hard Data solution: > Implemented strict RAG guardrails to ensure the AI distinguishes between estimated values and verified data.

- title: Distinguishing Materials
  solution: >
    Improved prompts to detect recycling codes, labels, and textual cues
    to differentiate visually similar materials like PLA and PET.

- title: Latency
  solution: >
    Optimized performance by running vision classification and
    vector retrieval in parallel threads.

accomplishments: - Choose_Better_Engine: > A recommendation system that provides nearby or similar sustainable alternatives instead of generic judgments.

- Multimodal_Granularity: >
    Successfully extracted small, dense text such as ingredient lists
    from user-uploaded product images.

- UI_UX_Simplification: >
    Converted complex carbon data into a simple Red/Yellow/Green
    traffic-light system understandable by all age groups.

learnings: - sustainability_is_not_binary: > Learned to model trade-offs and explain nuances such as recyclability versus transport emissions.

- agentic_ai_advantage: >
    Agentic AI with tools behaves more like a researcher than a chatbot.

- vision_as_input: >
    Visual input removes friction when users don’t know product names,
    making scanning the most natural interface.

next_steps: - AR_Overlay: > Real-time augmented reality overlays showing eco indicators in physical retail environments.

- Marketplace_Integration: >
    Enable one-click purchasing of better alternatives
    through sustainable brand partnerships.

- Gamification: >
    Introduce a Carbon Wallet with points and badges
    for CO2 savings.

- Enterprise_API: >
    Provide APIs for e-commerce platforms to auto-tag
    products with Eco Scores.

Built With

• gradio
• llamaindex
• python