Project Info

PM.ai

Inspiration

As engineers, we’ve all experienced the chaos of collaborating on complex projects—messages scattered across Discord, Slack, and Notion; roles unclear; and deadlines slipping through the cracks. During the hackathon, our team realized that what slows teams down isn’t just lack of talent—it’s lack of clarity. We wanted to build an AI-powered orchestrator that could read a project brief, understand what needs to be done, and automatically divide it into actionable tasks and clear roles — giving every team member instant alignment.

What it does

PM.ai is an intelligent project management assistant that: Reads your project or assignment brief (PDF/text) Automatically identifies roles (like Backend Developer, PM, UI/UX Designer) Breaks down large goals into actionable tasks and subtasks Assigns responsibilities based on role specialization Stores everything in a collaborative Supabase dashboard so teams can interact, claim tasks, and stay updated It acts like your AI project manager — one that never forgets, never misses a deadline, and keeps everyone on the same page.

How we built it

Our stack was designed around reliability, reasoning, and speed: Claude 4.5 (Anthropic) for advanced reasoning, structured JSON task generation, and long-context understanding of project briefs LangChain for chaining AI prompts and embedding logic Chroma for vector-based “memory” retrieval of project content Supabase for relational storage of roles, tasks, and subtasks (pgvector used for memory embeddings) FastAPI as the backend for uploading files, calling agents, and connecting to Supabase Supabase Edge Functions to bulk-insert and manage data asynchronously The workflow: User uploads a project brief → extracted with pdf-parse Text is chunked, embedded, and stored in Chroma Claude’s PM Agent generates high-level tasks and roles Claude’s Specialist Agents expand each task into technical subtasks 5.All data syncs into Supabase and is displayed in the interactive dashboard

Challenges we ran into

Model formatting: Early versions of Claude’s JSON responses were inconsistent. We had to sanitize and validate outputs before inserting them into Supabase. Supabase schema setup: Managing relationships between projects, tasks, and subtasks without breaking function permissions required careful design.

Accomplishments we're proud of

Built a fully functional multi-agent orchestration system that dynamically parses, structures, and assigns project tasks. Integrated Claude 4.5, Supabase, and Chroma successfully — all talking to each other through FastAPI. Achieved structured, reusable JSON pipelines for consistent AI task generation. Created a system that we actually used to manage our own hackathon workflow — proving its effectiveness. Built a team that combined AI, backend, and UX thinking seamlessly.

What we learned

The power of AI orchestration — how chaining reasoning agents can create structured workflows from unstructured input. How to design robust API systems that handle dynamic AI responses (even malformed JSON). The value of clear schema design in Supabase for linking project hierarchies (projects → tasks → subtasks). That collaboration and communication—ironically, the problem we were solving—are what make hackathon success possible.

What's next

Autonomous Planning: Let PM.ai dynamically replan deadlines or reassign roles when someone misses a milestone. SaaS Launch: Turn this into a lightweight productivity platform for startups, research labs, and university teams.

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only

CSSIn code
HTMLIn code
JavaScriptIn code
ReactIn code
SupabaseIn code
Tailwind CSSIn code
TypeScriptIn code
FastAPIClaimed
LangChainClaimed

7 of 9 appear in the indexed code. 2 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

No AI coding agent signals were found in this repository.

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

232 KB

Source files

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.