Project Description
This project is a green-field build of a Python backend that combines FastAPI with LangChain and LangGraph to orchestrate large-language-model workflows. The service must ingest raw data, run processing and analytical steps, expose secure endpoints, and stream results to clients in real time—all while running natively on Azure.
Core scope
• Data processing & analysis: structured and unstructured data land in Azure Storage, then flow through LangChain/LangGraph pipelines for enrichment and insight generation.
• User authentication & authorization: JWT-based security (Azure AD B2C or similar) on every route, with role-based access rules baked in.
• Real-time updates & notifications: WebSocket or Server-Sent Events layer so dashboards receive instant output as soon as the pipeline publishes.
Cloud architecture
– Azure Functions for lightweight event triggers (e.g., blob-created, queue messages).
– Azure Kubernetes Service to host the FastAPI app, LangChain agents, and any background workers.
– Azure Storage (Blob + Table) for raw files, intermediate artifacts, and metadata.
DevOps & delivery
The code sits in a Git repo with CI/CD that ships containers to AKS and auto-deploys Azure Functions. Infrastructure as Code (Bicep or Terraform) is required so every environment stays reproducible.
Acceptance criteria
1. Running AKS cluster with the FastAPI gateway reachable behind an Azure Application Gateway or Ingress.
2. End-to-end data flow demo: upload sample file → processing pipeline → real-time push to a subscribed client.
3. Auth flow proven with at least two user roles and protected endpoints.
4. Automated pipeline: `git push` to main triggers build, unit tests, container publish, and rollout.
5. Clear README and diagram explaining architecture, local dev steps, and production deployment commands.
Deliverables
• Full Python codebase (FastAPI, LangChain, LangGraph)
• IaC templates and pipeline YAML
• Test suite (unit + minimal integration)
• Deployment documentation
I’ll provide the Azure subscription, service principals, and any required model keys. Let’s build an efficient, scalable foundation that is easy to evolve as usage grows.