← All projects
active

Agentic SDLC Framework

Convention-driven PRD-to-HLD-to-LLD pipeline using schema-gen as a data model spine, context manifests in tickets, and human checkpoints between execution layers.

Agentic AIPythonDeveloper Tools

The Problem

AI-assisted code generation is powerful but chaotic. Without structure, LLM-generated code drifts from requirements, accumulates inconsistencies across files, and loses context between generation steps. The gap isn’t the AI’s coding ability — it’s the lack of a data model that connects product requirements through to implementation decisions.

Approach

The Agentic SDLC Framework introduces a convention-driven pipeline with three layers, each generating structured artifacts that feed the next:

  1. Schema-Gen — the foundational commit. Before any business logic, the AI generates Pydantic models, database schemas, and API contracts. This becomes the “spine” that every subsequent layer references.
  2. PRD → HLD → LLD Pipeline — Product requirements flow through high-level design to low-level design, with each stage producing a structured document (not free-form text). Context manifests link decisions across layers so the AI can trace why a particular implementation choice was made.
  3. Human Checkpoints — between every layer transition, a human reviews and approves. The AI proposes, the human disposes. This prevents the compound error problem where a wrong assumption in the PRD cascades through to broken code.

Each artifact is a versioned file in the repository, not ephemeral chat history. This means you can diff designs, roll back decisions, and onboard new team members by reading the pipeline output.

Current State

Active framework with a reference implementation (Todo app demonstrating the full pipeline). Being used to build TradeForge components. The schema-gen-first pattern has proven particularly effective — it eliminates an entire class of “the AI generated code that doesn’t match the data model” bugs.