Zenith

Zenith is a project management app built by a team of three developers. A manager writes a requirement, and AI agents break it into tasks and assign each one to the employee whose skills, workload and cost fit it best. I built its landing page, employee dashboard and setup flow, and the services that read and write its tasks and users in MongoDB.

Planning work across a team usually means a manager reading every requirement, splitting it into tasks by hand and guessing who has the time and the skills for each one. The tools for it are spread across trackers and spreadsheets, and nothing records why a task went to whom. Zenith hands the splitting and the matching to AI agents and keeps a log of every decision.

• →The landing page: the navbar, the hero with a rotating headline and an animated light-beam background, features, how it works, the tech stack, testimonials, the call to action and the footer.
• →The employee dashboard: its first version, with task totals, a productivity figure, search, filters by status and priority, and status updates.
• →The setup form: where a company creates its organisation and its first admin and picks a plan.
• →The data services: the task and user services that read and write MongoDB for those screens.

• →A manager uploads a requirement, and the agents decide whether it is a simple task or a feature that needs planning.
• →A product-manager agent and an architect agent break a feature into tasks with priorities and estimated hours.
• →An allocation agent assigns each task by matching it to employees' skills with embeddings, then weighing their workload, seniority and hourly cost.
• →Each assignee gets an email, and the results are saved for the manager's allocation log.
• →Separate dashboards for admins, managers and employees, and an assistant that answers questions and opens the right page.

• →Simple work skips the planning. The agent service first rates a requirement's complexity. A simple task goes straight to allocation, and only a feature goes through the product-manager and architect agents, so small requests take one model call instead of three.
• →Skills are matched by meaning, not by keyword. Employees' skills are indexed as embeddings in FAISS, so a task about "REST endpoints" can find someone whose skills say "API design".

The Next.js app serves the pages and reads and writes MongoDB through server actions. A manager's requirement goes to the agent service, a LangGraph workflow in Python that plans the work, assigns it and saves the result back to MongoDB, where the dashboards read it.

• →Live at zenith0.vercel.app.