llm_analysis.py

"""
PHASE 6: LLM Analysis via Groq API
Model: llama3-70b-8192

Rules:
- LLM ONLY explains and recommends
- LLM does NOT calculate eco score
- LLM does NOT predict impact category
- API key read from environment (HF Spaces Secret: GROQ_API_KEY)
"""

import os


def get_groq_client():
    try:
        from groq import Groq
        api_key = os.getenv("GROQ_API_KEY", "").strip()
        if not api_key:
            return None, "GROQ_API_KEY not set."
        return Groq(api_key=api_key), None
    except ImportError:
        return None, "groq package not installed."


def generate_eco_explanation(
    product_name: str,
    category: str,
    deforestation_risk: float,
    pollution_level: float,
    biodiversity_impact: float,
    eco_score: float,
    predicted_impact: str,
    packaging_type: str = "Unknown",
    ingredients: str = "Not specified",
) -> dict:
    """
    Generate LLM environmental explanation.
    LLM receives pre-computed score and label — it never recalculates them.
    """
    client, error = get_groq_client()
    if client is None:
        return _fallback(
            product_name, category, deforestation_risk,
            pollution_level, biodiversity_impact,
            eco_score, predicted_impact, error,
        )

    prompt = f"""You are an environmental scientist providing educational insights.

Product Details:
- Name: {product_name}
- Category: {category}
- Packaging: {packaging_type}
- Ingredients/Materials: {ingredients}
- Eco Score: {eco_score}/10  ← already calculated, do NOT recalculate
- Impact Category: {predicted_impact}  ← already predicted by ML, do NOT re-predict

Environmental Risk Metrics (0–1 scale):
- Deforestation Risk: {deforestation_risk}
- Pollution Level: {pollution_level}
- Biodiversity Impact: {biodiversity_impact}

Respond with EXACTLY these four sections:

## Environmental Impact Summary
[2–3 sentences explaining the product's overall environmental footprint in plain language]

## Deforestation Risk Analysis
[Explain what drives this product's deforestation risk, which regions/forests are affected, and why]

## Biodiversity & Species Impact
[Explain which wildlife and ecosystems are at risk; estimate how many species could be affected and why]

## 3 Eco-Friendly Alternatives
1. **[Name]**: [Why it's better and where to find it]
2. **[Name]**: [Why it's better and where to find it]
3. **[Name]**: [Why it's better and where to find it]

## Quick Sustainability Tips
[2–3 actionable tips the consumer can apply immediately]

Keep language accessible. Be factual and educational, not alarmist.
Do NOT produce any numerical eco score or impact category prediction."""

    try:
        response = client.chat.completions.create(
            model="llama-3.3-70b-versatile",
            messages=[
                {
                    "role": "system",
                    "content": (
                        "You are an expert environmental scientist. Provide factual, "
                        "accessible sustainability insights. Never recalculate scores "
                        "or re-predict categories — those come from separate systems."
                    ),
                },
                {"role": "user", "content": prompt},
            ],
            max_tokens=1200,
            temperature=0.7,
        )
        return {
            "success":     True,
            "explanation": response.choices[0].message.content,
            "model":       "llama-3.3-70b-versatile (Groq)",
            "tokens_used": response.usage.total_tokens,
        }
    except Exception as e:
        return _fallback(
            product_name, category, deforestation_risk,
            pollution_level, biodiversity_impact,
            eco_score, predicted_impact, str(e),
        )


def _fallback(
    product_name, category, deforestation_risk,
    pollution_level, biodiversity_impact,
    eco_score, predicted_impact, error_msg="",
) -> dict:
    """Template explanation used when Groq API is unavailable."""

    if deforestation_risk > 0.5:
        defo = (
            f"{product_name} carries a high deforestation risk ({deforestation_risk:.0%}). "
            "Supply chains likely involve palm oil, wood pulp, or agricultural land clearing "
            "in tropical regions such as the Amazon, Southeast Asian rainforests, or the Congo Basin."
        )
    elif deforestation_risk > 0.2:
        defo = (
            f"{product_name} has a moderate deforestation risk ({deforestation_risk:.0%}). "
            "Some raw materials may be linked to land-use change, though responsible sourcing "
            "certifications (FSC, RSPO) can mitigate this."
        )
    else:
        defo = (
            f"{product_name} has a low deforestation risk ({deforestation_risk:.0%}). "
            "Materials show minimal connection to forest loss, especially if recycled or certified organic."
        )

    if biodiversity_impact > 0.5:
        bio = (
            "High biodiversity impact. Estimated 50–200+ species in affected ecosystems "
            "face habitat disruption, including pollinators, soil microbiota, and apex predators. "
            "Pollution runoff may further threaten aquatic biodiversity."
        )
    elif biodiversity_impact > 0.2:
        bio = (
            "Moderate biodiversity impact. Localised ecosystems may experience disruption; "
            "approximately 10–50 species could be indirectly affected through habitat "
            "fragmentation and chemical pollution."
        )
    else:
        bio = (
            "Low biodiversity impact. Minimal disruption to local ecosystems. "
            "The product lifecycle has limited spillover effects on wildlife habitats."
        )

    alts_map = {
        "Shampoo": [
            "**Ethique Shampoo Bar** — Zero-plastic, concentrated, biodegradable",
            "**Briogeo Be Gentle** — Sulfate-free, plant-based ingredients",
            "**Baking Soda & Apple Cider Vinegar** — DIY ultra-low-impact alternative",
        ],
        "Snacks": [
            "**Local Organic Produce** — Minimal packaging, zero transport emissions",
            "**Bulk-store Nuts & Seeds** — Bring your own container, zero waste",
            "**Homemade Granola** — Full control over ingredients and packaging",
        ],
        "Plastic": [
            "**Stainless Steel Alternatives** — Reusable, durable, infinitely recyclable",
            "**Compostable Bioplastics (PLA)** — Breaks down in industrial composting",
            "**Glass Containers** — Inert, infinitely recyclable, no chemical leaching",
        ],
        "Cosmetics": [
            "**ILIA Beauty** — Certified organic, recycled packaging",
            "**RMS Beauty** — Raw food-grade ingredients, glass packaging",
            "**DIY Natural Cosmetics** — Beeswax, coconut oil, natural pigments",
        ],
        "Clothing": [
            "**Patagonia** — Recycled materials, repair programme, lifetime guarantee",
            "**Secondhand / Thrift** — Zero new production, circular economy",
            "**Tentree or Eileen Fisher** — Certified organic fibres, ethical supply chains",
        ],
    }
    alts = alts_map.get(category, [
        "**Local / Organic alternatives** — Reduced transport and chemical footprint",
        "**Secondhand or refurbished** — Circular economy approach",
        "**DIY or zero-waste options** — Maximum control over environmental impact",
    ])

    explanation = f"""## Environmental Impact Summary
{product_name} is a {category} product with an eco score of {eco_score}/10 ({predicted_impact}). \
{"This product raises significant environmental concerns — consider switching to one of the alternatives below." if eco_score < 5 else "This product has moderate to good sustainability credentials, though improvement is always possible."}

## Deforestation Risk Analysis
{defo}

## Biodiversity & Species Impact
{bio}

## 3 Eco-Friendly Alternatives
1. {alts[0]}
2. {alts[1]}
3. {alts[2]}

## Quick Sustainability Tips
- Look for certified eco-labels: FSC, Fair Trade, USDA Organic, or B Corp.
- Choose concentrated formulas and refillable containers to reduce packaging waste.
- Research brand sustainability reports and third-party certifications before purchasing."""

    note = f"Groq API unavailable: {error_msg}" if error_msg else "Using template explanation."
    return {
        "success":     True,
        "explanation": explanation,
        "model":       "Template Fallback",
        "note":        note,
    }


def compare_products_llm(product1: dict, product2: dict) -> dict:
    """LLM comparison explanation for two products (scores pre-calculated)."""
    client, error = get_groq_client()
    if client is None:
        winner = product1["name"] if product1.get("eco_score", 5) >= product2.get("eco_score", 5) else product2["name"]
        return {
            "success":    True,
            "comparison": f"**{winner}** is the more environmentally friendly choice based on its lower impact scores.",
            "model":      "Fallback",
        }

    prompt = f"""Compare these two products' environmental impact in 3–4 sentences.

Product A: {product1['name']} ({product1.get('category','')})
- Eco Score: {product1.get('eco_score','N/A')}/10
- Deforestation: {product1.get('deforestation_risk',0):.2f} | Pollution: {product1.get('pollution_level',0):.2f} | Biodiversity: {product1.get('biodiversity_impact',0):.2f}

Product B: {product2['name']} ({product2.get('category','')})
- Eco Score: {product2.get('eco_score','N/A')}/10
- Deforestation: {product2.get('deforestation_risk',0):.2f} | Pollution: {product2.get('pollution_level',0):.2f} | Biodiversity: {product2.get('biodiversity_impact',0):.2f}

State which is more sustainable, explain the key environmental trade-offs, and give a consumer recommendation.
Do NOT recalculate scores."""

    try:
        resp = client.chat.completions.create(
            model="llama-3.3-70b-versatile",
            messages=[{"role": "user", "content": prompt}],
            max_tokens=400,
            temperature=0.6,
        )
        return {
            "success":    True,
            "comparison": resp.choices[0].message.content,
            "model":      "llama-3.3-70b-versatile (Groq)",
        }
    except Exception as e:
        return {"success": False, "comparison": "Comparison unavailable.", "error": str(e)}