Custom GPT-style agent that helps procurement professionals design better solicitations using the PA Procurement Handbook to recommend creative but compliant procurement strategies.
DRAFT2 weeks (MVP)
$1
0 features
0 milestones
Project Overview
Problem Statement
Procurement professionals often default to familiar procurement strategies, leading to narrow thinking and mismatched procurement methods (“round peg in a square hole”). The tool expands the art of the possible by walking users through all compliant options and guiding them to the best-fit, handbook-aligned solicitation strategy.
Ideal Customer Profile
Primary: Pennsylvania Commonwealth procurement professionals and program leads (issuing officers, contracting staff, program SMEs). Includes policy, legal, BDISBO, and other relevant internal stakeholders who support procurement strategy and solicitation development. Secondary: learners/stakeholders interested in understanding the procurement process.
App Audience
Internal tool for Commonwealth users (with optional read-only/educational mode for broader audiences).
Project Type
Greenfield (Custom GPT in ChatGPT)
N/A
Integrations
Generate Word solicitation templates for export (likely via GPT Actions); coordinate with AI lead (Harrison McRae) and OA/OA-OIT as needed. Other integrations TBD.
Tech Stack Preferences
Custom GPT in ChatGPT; consider GPT Actions + document generation service for Word (.docx) exports; final stack TBD with OA/AI lead.
Deployment Preferences
Hosted within ChatGPT as a Custom GPT; internal access via ChatGPT org/workspace; any Actions hosted in Commonwealth-approved environment.
Design Status
No separate UI designs expected for MVP; rely on ChatGPT Custom GPT interface; include export flow for Word documents.
Open Questions
Should the PRD include a budget/milestone payment structure even though development is covered by an existing contract? If not, what internal milestone checkpoints should we use instead?
User indicated they already have a developer on contract and doesn't want to include budget info. PRDAgent workflow requires milestones; we can set payments to $0 or treat milestones as internal checkpoints.
Do we need to fine-tune/train a model on the handbook content, or use retrieval (RAG) via Custom GPT Knowledge files? If training is required, what platform/approval process will OA/OA-OIT require?
User requested 'training the foundation model on all chapters.' In ChatGPT Custom GPT, typical approach is Knowledge + instructions; fine-tuning/training may require separate infrastructure and approvals. Coordination mentioned with AI lead Harrison McRae and OA.