APEX Lite is an AI-powered project planning system designed to automatically generate structured project management artifacts from simple natural language inputs. The system operates entirely offline using a locally hosted small-parameter Large Language Model (Phi) via Ollama, ensuring data privacy and eliminating dependency on cloud-based APIs. This research focuses on evaluating the impact of prompt engineering strategies on the quality of AI-generated project artifacts. Two prompting approaches—Role Prompting and Structured Prompting—are systematically compared across 20 diverse project scenarios. The system generates key artifacts including:- Business Requirements Document (BRD)- Work Breakdown Structure (WBS)- Risk Register- Cost Estimation- Agile Backlog Outputs are evaluated using three metrics:- Completeness- Structural Consistency- Clarity Experimental results demonstrate that Structured Prompting significantly improves output structure and parsing reliability, while Role Prompting produces more coherent and descriptive outputs for narrative artifacts such as BRDs. A hybrid prompting strategy is therefore recommended for optimal performance. The system is implemented using Python, Streamlit, Pandas, Plotly, and python-docx, with modular architecture enabling scalable extension to additional planning components. This work highlights that effective prompt engineering, rather than model scale alone, is a critical factor in achieving reliable AI-assisted project planning using local LLMs.This study has several limitations. Evaluation scores were assigned by a single reviewer, introducing potential subjectivity. The dataset of 20 project scenarios is relatively small. Additionally, results are specific to the Phi-2 model and may not generalize to larger or alternative LLMs. Keywords: Prompt Engineering, Local LLM, Project Management Automation, BRD, WBS, Ollama, Phi Model, Streamlit
Sugnik Tarafder (Sun,) studied this question.