Skip to main content

Tech Stack

The AI Ingredient Scanner leverages modern AI and cloud technologies to deliver reliable ingredient analysis.

Architecture Diagram​

Technology Stack​

Core AI​

TechnologyPurposeDetails
Google Gemini 2.0 FlashLLMAnalysis, validation, translation
LangGraphOrchestrationMulti-agent workflow management
LangSmithTracingLLM call logging and debugging

Backend​

TechnologyPurposeDetails
Python 3.11+LanguageType hints, async support
FastAPIREST APIMobile app integration
StreamlitWeb UIInteractive dashboard
PydanticValidationRequest/response schemas

Data Layer​

TechnologyPurposeDetails
Qdrant CloudVector DBSemantic ingredient search
Redis CloudCachingSession persistence
Google EmbeddingsVectorsgemini-embedding-001 model

Mobile​

TechnologyPurposeDetails
React NativeFrameworkCross-platform mobile
ExpoToolchainDevelopment and builds
TypeScriptLanguageType-safe mobile code
expo-cameraCameraNative camera access

LLM Integration​

Gemini 2.0 Flash​

Used for all AI capabilities:

# Analysis generation
client.models.generate_content(
    model="gemini-2.0-flash",
    contents=analysis_prompt,
)

# Vision OCR with translation
client.models.generate_content(
    model="gemini-2.0-flash",
    contents=[image_part, ocr_prompt],
)

# Google Search grounding
config = types.GenerateContentConfig(
    tools=[types.Tool(google_search=types.GoogleSearch())]
)

Model Selection Rationale​

ModelUse CaseWhy
gemini-2.0-flashAll operationsFast, capable, cost-effective
gemini-embedding-001Vector searchHigh-quality embeddings

Vector Database​

Qdrant Configuration​

COLLECTION_NAME = "ingredients"
VECTOR_SIZE = 768  # gemini-embedding-001 output
DISTANCE = Distance.COSINE

# Query
results = client.query_points(
    collection_name=COLLECTION_NAME,
    query=embedding,
    limit=1,
)

Payload Schema​

{
  "name": "Glycerin",
  "purpose": "Humectant, moisturizer",
  "safety_rating": 9,
  "concerns": "No known concerns",
  "recommendation": "SAFE",
  "allergy_risk_flag": "low",
  "origin": "Natural",
  "category": "Both",
  "regulatory_status": "FDA approved, EU compliant"
}

API Communication​

Environment Configuration​

# Required
GOOGLE_API_KEY=         # Gemini API access
QDRANT_URL=             # Vector database
QDRANT_API_KEY=         # Qdrant authentication

# Optional
REDIS_URL=              # Session caching
LANGCHAIN_API_KEY=      # LangSmith tracing

Deployment Options​

EnvironmentStack
Local DevelopmentStreamlit + uvicorn
Production APIGoogle Cloud Run
Mobile TestingExpo Go
Production MobileEAS Build

Performance Optimization​

Caching Strategy​

  • Qdrant: Ingredient data persisted permanently
  • Redis: Session state cached for 24 hours
  • LRU Cache: In-memory settings caching

Parallel Processing​

# Research agent processes 3 ingredients per worker
BATCH_SIZE = 3

with ThreadPoolExecutor(max_workers=num_workers) as executor:
    futures = {
        executor.submit(_research_batch, idx, batch): idx
        for idx, batch in enumerate(batches)
    }