Ai document generator

Project Overview

I designed and built an AI-powered Document Engineering Platform that transforms raw, unstructured user input into production-ready software engineering deliverables, including:

Software Requirements Specification (SRS)

Technical Proposal

High-Level Design (HLD)

Low-Level Design (LLD)

The platform acts as an AI-assisted architecture workspace for solution architects and engineering teams, enabling them to move from vague ideas to structured, enterprise-grade documentation through an intelligent, iterative workflow.

At its core, the system combines LLM orchestration, structured generation pipelines, and human-in-the-loop approval flows to ensure both speed and accuracy.

My Role

I led the end-to-end system design and implementation, including:

AI workflow architecture (n8n + LLM orchestration)

Prompt engineering and structured generation design

Backend pipeline design and document lifecycle management

Frontend integration strategy (interactive SRS workspace)

System scalability and production-readiness decisions

This was built as a production-grade system, not a prototype.

Architecture & Approach

The system follows a pipeline-driven, backend-first architecture:

Input ingestion (messy or structured requirements)

AI classification & normalization

Structured document generation (modular sections)

Human approval & iteration loop

Multi-document transformation pipeline (SRS → Proposal → HLD → LLD)

Final document export (DOCX/PDF)

A key design principle was strict structure enforcement with flexible user interaction, ensuring outputs remain consistent while still editable.

Tech Stack & Frameworks

Frontend: React (Lovable platform for rapid UI prototyping)

Backend / Orchestration: n8n (self-hosted workflow engine)

AI Models: Google Vertex AI (Gemini)

Cloud: Google Cloud Platform

Document Processing: Custom DOCX generation pipelines

Data Handling: JSON-based structured schemas

Key Engineering Contributions

1. AI-Powered Requirements Understanding

Designed a classification and transformation system that can handle multiple input types:

Vision-only ideas

Raw feature lists

Semi-structured requirements

Enterprise-level requirement dumps

Each input is automatically classified and expanded into a complete SRS baseline, including:

Functional Requirements (FR)

Non-Functional Requirements (NFR)

Scope / Out of Scope

Assumptions, Constraints, Risks

2. Structured Document Generation Pipeline

Built a modular LLM generation system where:

Each section (e.g., Scope, FRs, Risks) is generated independently

Outputs follow strict JSON schemas

Sections are merged into consistent, professional documents

This ensures:

High reliability

Easy debugging and regeneration

Section-level editing and refinement

3. Human-in-the-Loop Approval System

Implemented an interactive document workspace with:

Section-level approval / de-approval

AI-assisted refinement per section

Expand / collapse controls for large documents

Controlled regeneration without breaking structure

This bridges the gap between AI automation and real engineering workflows.

4. Multi-Stage Engineering Pipeline

Designed a progressive document pipeline:

SRS → Technical Proposal → HLD → LLD

Each stage:

Uses the previous document as input

Refines and expands system detail

Maintains cross-document consistency

This creates a full engineering lifecycle system, not just a generator.

5. Prompt Engineering & Output Control

Developed production-grade prompt strategies to ensure:

Strict JSON-only outputs (no hallucinated structure)

No unintended feature additions

Consistent terminology and tone

Alignment with real-world engineering standards

Also implemented:

Section-aware prompting

Context injection (cross-section awareness)

Safe refinement rules (no breaking changes)

6. Document Export & Professional Formatting

Built a system to generate client-ready deliverables:

Structured DOCX export

Consistent formatting across sections

Tables for FRs, NFRs, and risks

Enterprise-ready presentation (branding-ready)

Key Strengths of the System

AI + workflow orchestration (not just prompting)

Strong separation of structure, logic, and presentation

Handles messy real-world input (not ideal inputs only)

Human-in-the-loop validation (critical for enterprise use)

Fully extensible for additional document types or domains

Outcome

The platform enables teams to:

Convert vague ideas into structured engineering artifacts

Accelerate solution architecture and proposal creation

Maintain consistency across all design documents

Reduce time from concept to implementation planning

It effectively acts as a design accelerator for engineering teams, combining AI speed with human control.