How rajeshbakthavachalam/agenticsdlc works
Project type and purpose
• AgenticSDLC is a full-stack, AI-driven SDLC orchestration platform tailored to automate the Software Development Lifecycle of complex projects (in this example, a WiFi7 management platform).
• It exposes both a programmatic REST API (FastAPI) and an interactive UI (Streamlit). Internally it uses a graph-based workflow engine, integrates with external LLM services, persists state in Redis, and emits Markdown artifacts to disk.
File structure
• Top-level entry points:
• app_api.py – FastAPI server
• app_streamlit.py – Streamlit UI
• dockerfile – containerizes Redis + application
• requirements.txt – dependencies
• Core code under src/dev_pilot:
• api/fastapi_app.py – API routing and controllers
• ui/streamlit_ui/streamlit_app.py – UI components and controls
• graph/graph_builder.py & graph_executor.py – constructs and runs the LangGraph workflow
• LLMS/… – adapters to Gemini, Groq, OpenAI
• cache/redis_cache.py – Redis connection and state persistence
• nodes/… – individual nodes for each SDLC phase (requirements, design, code, tests, etc.)
• state/sdlc_state.py – data models for session state
• dto/… – request/response schemas for API
• utils/… – logging, constants, helpers
Key insights from README
• Graph‐based workflow (LangGraph) drives sequential SDLC phases.
• Human-in-the-loop approval loops at each node.
• Generates artifacts to “artifacts/” (Markdown files).
• External services: LLM APIs and Redis.
• Two frontends: Streamlit (web UI) and FastAPI (REST).
Main components & relationships
A. Frontend layer
• Streamlit UI ──> calls FastAPI endpoints OR invokes graph executor directly
• REST API (FastAPI) ──> entrypoint for programmatic access
B. Backend layer
• Controller/service layer (api/) receives requests, transforms into DTOs, delegates to workflow engine
• Workflow engine (graph_builder + graph_executor) orchestrates nodes in sequence
• Node modules (nodes/*) implement each SDLC step, interacting with LLM clients for artifact generation
• State management (redis_cache + sdlc_state) for persisting session progress
C. External services
• LLM providers (OpenAI, Gemini, Groq) called by LLMS adapters
• Redis – key/value store for session state
D. Artifacts storage
• Filesystem “artifacts/” – stores generated Markdown docs per phase
Architectural patterns & design principles
• Modular, plugin-style nodes for each SDLC phase (Strategy/Command pattern)
• Graph orchestration (Workflow engine pattern)
• Clear separation: UI vs API vs core engine vs persistence vs external integrations
• Dependency injection of LLM clients & cache to enable easy swapping/testing
Final checklist
• Clear, descriptive labels on all components
• A note on the modular node structure and pluggable LLM adapters