TimeWeaver

AI-native task manager — natural language in, structured plan out.

The Brief · Design Challenge
"How might I build a task manager that works the way humans actually think — in fragments, in natural language, without structure upfront?"

Design Principles

Natural language first

No forms, no dropdowns — type how you think.

Zero friction to start

Useful within 30 seconds of first open.

AI as collaborator, not gimmick

Gemini earns its place in every interaction.

3 → 9 / 10

AI confidence

20+

Prompt structures tested

Live on Vercel

Deployment

End-to-end

Built solo

Key Challenges

User Challenges

1.People abandon task apps when they require too much setup.
2.Natural language is ambiguous — AI interpretation had to be reliable.
3.The product needed to feel useful within 30 seconds of first use.

Design & Constraint Challenges

1.Prompt engineering is both a design and an engineering problem — had to learn both.
2.Migrating from Streamlit to Vercel mid-project required rebuilding deployment from scratch.
3.No team, no feedback loop — had to self-test rigorously.

image placeholder

Research artifact

Drop in an affinity diagram, journey map, or synthesis board that shaped the direction.

Situation

Existing task tools require rigid structure upfront — causing abandonment. A clear gap existed for an AI-native task manager that parses natural language and organizes intelligently.

Task

For a graduate seminar with Professor Ronald Beghetto, I independently designed, built, and deployed TimeWeaver — an AI-powered task management chatbot — going from zero coding confidence to a fully deployed product.

Action

Integrated the Google Gemini API for natural-language task parsing — users type how they think, Gemini structures it.
Built v1 in Streamlit, then migrated the entire app to Vercel for production — learned deployment pipelines from scratch.
Designed a conversation UI with zero forms or dropdowns — pure natural language.
Iterated prompt engineering across 20+ prompt structures to improve categorization accuracy.
Grew self-rated AI confidence from 3/10 → 9/10 through this project.

Result

Fully deployed live product on Vercel.
Gemini integration successfully parsed and categorized tasks from unstructured input.
Demonstrated end-to-end product ownership: research → design → build → deploy.
Proof of concept validated AI-native productivity tooling.

image placeholder

Process & exploration

Sketches, flow diagrams, wireframes, or iteration shots.

image placeholder

Final outcome

Hero shot of the final screen, artifact, or installed deliverable.

Tech Stack

PythonStreamlit → VercelGoogle Gemini APIGitHub

Research Methods

Prompt engineeringRapid prototypingSelf-directed learning

AI Used

Google Gemini 1.5 Pro — natural language task parsing