E-Comm POC - AI Customer Service & Content Generation

Technology Stack

AI Agent development projects featuring E-Commerce Customer Service with CrewAI RAG and LinkedIn Content Generation with Google ADK. Multi-agent orchestration for real-world business automation.

OrchestrationCrewAI + Google ADK

LLMGPT-4o-mini + Gemini

Vector DBChromaDB

EmbeddingsOpenAI text-embedding-3

Image GenStability AI SDXL

CrewAI Customer Service Agents

🔍

Query Analyzer

Classify & Extract Intent

📚

Policy Retrieval

RAG Search + Eligibility

💬

Response Generator

Empathetic Response

Workflow 1: Customer Service Query Resolution

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#C17852', 'primaryTextColor': '#F0F6FC', 'primaryBorderColor': '#4A5E32', 'lineColor': '#E6C98F', 'secondaryColor': '#161B22', 'tertiaryColor': '#0D1117'}}}%%
flowchart LR
    subgraph Input["📥 CUSTOMER QUERY"]
        QUERY["💬 Raw Query"]
    end
    subgraph Agent1["🔍 QUERY ANALYZER"]
        CLASS["📋 Classify Request"]
        EXTRACT["🏷️ Extract Category"]
        TONE["😊 Detect Tone"]
    end
    subgraph Agent2["📚 POLICY RETRIEVAL"]
        RAG["🔎 RAG Search"]
        CHROMA[("🗄️ ChromaDB")]
        ELIG["✅ Eligibility Check"]
        SUM["📝 Summarize Policy"]
    end
    subgraph Agent3["💬 RESPONSE GENERATOR"]
        CRAFT["✍️ Craft Response"]
        STEPS["📋 Step-by-Step"]
    end
    subgraph Output["📤 OUTPUT"]
        RESP["💬 Customer Response"]
    end
    QUERY --> CLASS
    CLASS --> EXTRACT
    EXTRACT --> TONE
    TONE --> RAG
    RAG --> CHROMA
    CHROMA --> ELIG
    ELIG --> SUM
    SUM --> CRAFT
    CRAFT --> STEPS
    STEPS --> RESP

Workflow 2: RAG Document Ingestion Pipeline

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#C17852', 'primaryTextColor': '#F0F6FC', 'primaryBorderColor': '#4A5E32', 'lineColor': '#E6C98F', 'secondaryColor': '#161B22', 'tertiaryColor': '#0D1117'}}}%%
flowchart TB
    subgraph Input["📥 DOCUMENTS"]
        PDF["📄 PDF Files"]
        TXT["📝 TXT Files"]
        MD["📋 MD Files"]
    end
    subgraph Process["🔧 PROCESSING"]
        LOAD["📥 Load Documents"]
        CHUNK["✂️ RecursiveCharacterTextSplitter
1000 chars, 200 overlap"]
    end
    subgraph Embed["🧠 EMBEDDING"]
        EMB["🔢 text-embedding-3-small"]
        VEC["📊 Vector Embeddings"]
    end
    subgraph Store["💾 STORAGE"]
        CHROMA[("🗄️ ChromaDB
+ Metadata")]
    end
    PDF --> LOAD
    TXT --> LOAD
    MD --> LOAD
    LOAD --> CHUNK
    CHUNK --> EMB
    EMB --> VEC
    VEC --> CHROMA

Workflow 3: LinkedIn News Discovery

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#C17852', 'primaryTextColor': '#F0F6FC', 'primaryBorderColor': '#4A5E32', 'lineColor': '#E6C98F', 'secondaryColor': '#161B22', 'tertiaryColor': '#0D1117'}}}%%
flowchart LR
    subgraph Input["📥 TOPIC"]
        TOPIC["🤖 AI & ML Topic"]
    end
    subgraph Agent["🔍 NEWS RESEARCH AGENT"]
        SEARCH["🌐 Web Search"]
        FILTER["📰 Filter Sources
TechCrunch, ArXiv, VentureBeat"]
        CATEGORIZE["🏷️ Categorize
Launch, Research, Funding"]
    end
    subgraph Output["📤 RESULTS"]
        JSON["📄 JSON Array
Up to 5 Articles"]
    end
    TOPIC --> SEARCH
    SEARCH --> FILTER
    FILTER --> CATEGORIZE
    CATEGORIZE --> JSON

Workflow 4: LinkedIn Post Generation with Humor

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#C17852', 'primaryTextColor': '#F0F6FC', 'primaryBorderColor': '#4A5E32', 'lineColor': '#E6C98F', 'secondaryColor': '#161B22', 'tertiaryColor': '#0D1117'}}}%%
flowchart TB
    subgraph Input["📥 NEWS ARTICLE"]
        ARTICLE["📰 Article Data"]
    end
    subgraph Generate["✍️ POST GENERATOR"]
        HOOK["🎣 Hook"]
        CONTEXT["📝 Context"]
        INSIGHT["💡 Insight"]
        CTA["📢 Call-to-Action"]
    end
    subgraph Humanize["😄 HUMANIZER AGENT"]
        PERSONALITY["🎭 Add Personality"]
        HUMOR["😂 Inject Humor
Scale: 0-10"]
        TONE["💬 Conversational Tone"]
    end
    subgraph Loop["🔁 LOOP AGENT"]
        NEXT["📰 Next Article"]
    end
    subgraph Output["📤 OUTPUT"]
        POST["📝 LinkedIn Post
Max 1500 chars"]
    end
    ARTICLE --> HOOK
    HOOK --> CONTEXT
    CONTEXT --> INSIGHT
    INSIGHT --> CTA
    CTA --> PERSONALITY
    PERSONALITY --> HUMOR
    HUMOR --> TONE
    TONE --> POST
    POST --> NEXT
    NEXT --> ARTICLE

Workflow 5: AI Image Generation for Posts

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#C17852', 'primaryTextColor': '#F0F6FC', 'primaryBorderColor': '#4A5E32', 'lineColor': '#E6C98F', 'secondaryColor': '#161B22', 'tertiaryColor': '#0D1117'}}}%%
flowchart LR
    subgraph Input["📥 POST BODY"]
        TEXT["📝 LinkedIn Post"]
    end
    subgraph Prompt["🎨 IMAGE PROMPT AGENT"]
        ENGINEER["✨ Prompt Engineering
Style, Composition, Lighting"]
    end
    subgraph Generate["🖼️ IMAGE GENERATOR"]
        API["🔌 Replicate API"]
        SDXL["🎨 Stability AI SDXL"]
        POLL["🔄 Poll Completion
60x @ 2s intervals"]
    end
    subgraph Output["📤 RESULT"]
        URL["🔗 Image URL"]
        META["📋 Metadata"]
    end
    TEXT --> ENGINEER
    ENGINEER --> API
    API --> SDXL
    SDXL --> POLL
    POLL --> URL
    URL --> META

Workflow 6: Post Quality Evaluation & Batch Export

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#C17852', 'primaryTextColor': '#F0F6FC', 'primaryBorderColor': '#4A5E32', 'lineColor': '#E6C98F', 'secondaryColor': '#161B22', 'tertiaryColor': '#0D1117'}}}%%
flowchart TB
    subgraph Input["📥 POST TEXT"]
        POST["📝 LinkedIn Post"]
    end
    subgraph Evaluate["📊 READABILITY EVALUATOR"]
        FLESCH["📈 Flesch Reading Ease
0-100"]
        SCALE["🎯 LinkedIn Scale
1-10"]
        ANALYZE["🔍 Sentence Complexity"]
    end
    subgraph Batch["💾 BATCH SAVE AGENT"]
        COMPILE["📦 Compile All Posts"]
        SAVE["💾 save_posts_to_json"]
    end
    subgraph Output["📤 EXPORT"]
        JSON[("📄 linkedin_posts_batch.json")]
        FIELDS["📋 URL, Title, Category
Body, Image, Score"]
    end
    POST --> FLESCH
    FLESCH --> SCALE
    SCALE --> ANALYZE
    ANALYZE --> COMPILE
    COMPILE --> SAVE
    SAVE --> JSON
    JSON --> FIELDS