🎉 Coming Soon: Fully operable Data Alchemist paltform!

Experience the full agentic workflow system with real-time API integration and live data processing. Till then , explore the architecture and share your feedback.

Agentic Workflow System

Multi-Agent Architecture for Intelligent Customer Support

Autonomous agents communicating through pub/sub messaging with LangGraph orchestration, Neo4j graph database, and Google Gemini AI

Interactive Architecture Simulation

Watch real-time data flow through our multi-agent system. Click any simulation button to see how requests are processed across different workflows.

PUBLIC USERBrowser / MobileHTTP RequestsREST API ClientLEADERSHIPAdmin DashboardAnalytics ViewMetrics & ReportsMASTER AGENTLangGraph OrchestratorPort 8000State ManagementPUB/SUBAsync Message BusPort 8001Event DistributionINGEST AGENTData ProcessingPort 8002PII RedactionCHAT AGENTRAG + SynthesisPort 8003Vector SearchANALYTICS AGENTMetrics AggregatorPort 8004Data InsightsNEO4J DBGraph DatabaseVector IndexKnowledge GraphGEMINI 2.5LLM ProcessingText GenerationEmbeddings
System Ready.

Agent Deep Dive

Explore each autonomous agent's role, workflow, and capabilities

Master Agent

Port 8000

API Gateway & LangGraph Orchestrator

Receives HTTP requests and orchestrates workflows using LangGraph state machines

Workflow

1

Receives client HTTP requests

2

Creates correlation IDs for tracking

3

Publishes requests to pub/sub topics

4

Manages state: route → wait_response → complete

5

Polls for responses using peek/acknowledge

6

Returns results to clients

Key Features

Stateful workflow management

Non-blocking async handling

Timeout management (300s)

Proxies analytics requests

Dependencies

FastAPI, LangGraph, httpx, structlog

Technology Stack

Production-ready technologies powering the agentic architecture

Core Framework

Python 3.11

Async/await for concurrent processing

FastAPI

High-performance async web framework

LangGraph

State machine workflow orchestration

Pydantic

Data validation and type safety

Data Layer

Neo4j 5.14+

Graph database with vector search

Vector Embeddings

768-dimensional semantic search

Cypher Query

Graph traversal and analytics

Async Driver

Non-blocking database operations

AI & ML

Google Gemini

LLM for extraction and synthesis

text-embedding-004

Embedding model (768-dim)

Cosine Similarity

Vector comparison for search

Composite Ranking

Multi-factor relevance scoring

Infrastructure

Docker

Containerization for each agent

Docker Compose

Local development orchestration

Cloud Run

Serverless container platform

GCP Secrets

Secure credential management

Messaging

Async Queues

In-memory pub/sub with deque

Correlation IDs

Request tracking across agents

Peek/Acknowledge

Reliable message delivery

Topic Routing

Event-driven communication

Security & Quality

PII Redaction

Automatic sensitive data removal

Structured Logging

Debugging with structlog

Type Safety

Full TypeScript/Pydantic coverage

Health Checks

Service monitoring endpoints

Why This Architecture?

📈

Independent Scaling

Each agent scales based on its workload (10 chat agents, 2 ingest agents)

🛡️

Fault Isolation

One agent failure doesn't affect others - graceful degradation

🔄

Zero Downtime

Update agents independently without system-wide redeployment

💰

Cost Efficient

Cloud Run scales to zero - pay only for actual usage

🧪

Easy Testing

Test agents in isolation with clear pub/sub contracts

🔌

Extensible

Add new agent types without modifying existing code

10-30s
Ingest Processing Time
5-15s
Chat Response Time
<100ms
Master Agent Overhead
<10ms
Pub/Sub Latency

Production-Ready Agentic System

Built for scale, reliability, and maintainability with modern cloud-native technologies

5 Independent Agents
Async Messaging
Vector Search
LangGraph Orchestration