Dietly — The Nutrition Copilot That Doesn't Waste Your Time

Dietly is an AI copilot for nutrition labels. Scan a product, get a verdict. No fluff.

The Problem with Nutrition AI

Most AI nutrition tools are annoyingly polite:

"This product contains sodium, which could potentially be a consideration for some individuals who may be monitoring their intake..."

That's useless. You want to know: Should I eat this or not?

What Dietly Does Different

Dietly gives you a verdict, not a lecture.

❌ "This is high in sodium which may affect some people"
✅ "800mg sodium = 35% of your daily limit. This will spike your blood pressure. AVOID if you have hypertension."

Core Principles

1. Decisive — Clear verdicts: RECOMMENDED / PROCEED WITH CAUTION / AVOID
2. Data-First — Every claim has a number. "6g saturated fat (30% daily limit)" not "high in fat"
3. Goal-Oriented — Analysis tied to YOUR intent, not generic health advice
4. No Filler — Zero "Great question!" or "I'd be happy to help"

Transparent Assumptions

The copilot tells you what it's assuming before acting:

🎯 "I'm assuming you're checking this for heart health"
   [protein source] [quick energy] [diet compliance]

• Agree? It proceeds on your behalf with focused analysis
• Disagree? Click an alternative — it re-analyzes instantly

No silent assumptions. No guessing games.

Knows Its Limits

When data is incomplete or context matters, it says so:

"I can only see 6 nutrients. Without fiber data, I can't 
assess digestive impact."

"For someone with hypertension, this is risky. For a 
healthy athlete post-workout, it's fine."

It doesn't pretend to know what it doesn't.

---

How It Works

Multi-Agent Pipeline

Your nutrition label goes through 4 specialized agents:

📸 Scan Label
     ↓
┌─────────────────────────────────────────────┐
│  PARSER AGENT                               │
│  Extracts nutrients from OCR text           │
├─────────────────────────────────────────────┤
│  INTENT AGENT                               │
│  "Why did you scan this?"                   │
│  → quick energy? protein source? diet check?│
├─────────────────────────────────────────────┤
│  FILTER AGENT                               │
│  Removes irrelevant nutrients               │
│  → Checking protein? Ignore vitamins.       │
├─────────────────────────────────────────────┤
│  ANALYSIS AGENT                             │
│  Verdict tied to YOUR goal                  │
│  → "For quick energy: NOT IDEAL. Too much   │
│     fat will slow digestion."               │
└─────────────────────────────────────────────┘
     ↓
📊 Verdict + Concerns + Recommendation

---

Tech Stack

Layer	Tech
AI	Gemini 2.5 Flash via Langchain
Backend	Express.js, Socket.io
Frontend	React 19, Vite, Tailwind
OCR	Tesseract.js

---

Run It

Prerequisites: Node.js 18+, Gemini API key

# 1. Clone and setup backend
cd be
npm install
echo "GEMINI_API_KEY=your_key_here" > .env
npm run dev

# 2. In another terminal, setup frontend
cd fe
npm install
npm run dev

Open http://localhost:5173 → Upload a nutrition label → Get verdict

---

API

Endpoint	What it does
POST /api/v1/scan/analyze/stream	Scan + full analysis (SSE)
POST /api/v1/scan/reanalyze	Re-analyze with different intent
POST /api/v1/chat	Follow-up conversation

---

Project Structure

be/src/agent/
├── parser-agent.js      # OCR → structured nutrients
├── intent-agent.js      # Infer user's goal
├── filter-agent.js      # Remove noise
├── analysis-agent.js    # Generate verdict
├── orchestrator-agent.js # Coordinate pipeline
└── copilot-agent.js     # Chat follow-ups

fe/src/pages/
├── IngredientScanner.jsx # Scan + results
└── Chat.jsx              # Follow-up chat

---

Example Output

Input: Burger nutrition label

Agent output:

🎯 Assuming: heart health check
📊 Focusing on 3 of 9 nutrients

⚠️ NOT IDEAL FOR YOUR GOAL

Saturated Fat: 6g (30% daily limit)
→ Directly raises LDL cholesterol

Cholesterol: 105mg (35% daily limit)  
→ Compounds the saturated fat issue

Sodium: 90mg (4% daily limit)
→ Actually good for a burger

VERDICT: The low sodium is a plus, but saturated 
fat and cholesterol make this risky for heart health.
Consider leaner alternatives.

---

What This Isn't

• ❌ A calorie counter
• ❌ A meal planner
• ❌ A polite assistant that hedges every answer

What This Is

• ✅ A copilot that tells you the truth
• ✅ Analysis tied to your specific goal
• ✅ Data-backed verdicts you can act on

---

Future

Better UI
Real-time typing effects, animated agent transitions, and a cleaner mobile experience.

Open Food Facts API
Compare scanned products against a database of 3M+ products. Get recommendations like "This has 40% less sugar than similar cereals."

LangGraph Integration
Migrate from sequential agent calls to LangGraph's state machine. Enables conditional branching, parallel agent execution, and more complex reasoning flows.

---

Built for Encode 2026