Docket

Inspiration

Docket was born from a personal journey. As a pre-law student, I spent semesters buried in syllabi—manually building course outlines and color-coding calendars just to stay afloat. I loved the law, took the LSAT, and seriously considered the path, but along the way I discovered a passion for computer science. That shift made me realize I could build the tools I had always wished existed.

When I teamed up with my partner, a CS major, we saw an opportunity to combine our strengths: firsthand understanding of law students’ needs with the technical ability to solve them. Docket sits at that intersection as a product shaped by lived experience and built with modern engineering.

What It Does

Docket is an AI-powered study planner that goes far beyond simple syllabus scraping.

Upload a syllabus PDF and Docket:

Extracts all deadlines and key events
Breaks assignments into sub-tasks
Generates load-balanced study sessions
Flags workload conflicts

It also deploys a resource agent that autonomously searches:

Web resources
YouTube
Academic platforms
Practice problem sites

The agent iterates until key topics are covered, producing a curated study toolkit.

To stay efficient on free-tier LLMs, Docket uses:

Regex-based preprocessing to remove boilerplate
Token-aware chunking for long PDFs

Additional features:

Structured course outline builder
Cold call tracker
Pomodoro timer
Auto-generated task checklists
Multi-course management
One-click Google Calendar export

How We Built It

We began with backend prototypes for syllabus parsing and front-end mockups in Figma. Using Figma annotations, we analyzed real syllabi and mapped where time was being lost—this directly informed our product design and automation strategy.

Tech stack:

Frontend: React, TypeScript, Vite, Tailwind CSS
Backend: FastAPI with async MongoDB (per-user persistence)
Auth: Firebase (Google Sign-In)

AI Pipeline:

Powered by Groq’s Llama 3.1
Three-stage agentic flow:
1. Extract syllabus events
2. Generate optimized study plan
3. Curate resources via a agent

We integrated the Tavily API to enable autonomous search across multiple platforms, looping until coverage is sufficient.

The UI was designed mobile-first with a vintage clerk’s desk aesthetic (cream + sage), based on Figma and Canva mockups.

Challenges We Ran Into

Storage & architecture:
Our initial data storage approach broke the upload flow, data cleared mid-process, causing blank screens. We rolled back and rebuilt using MongoDB, separating the upload pipeline from persistence.

Rate limits & performance:
Groq’s limits required:

15-second delays between pipeline steps
Careful chunking of large PDFs
Efficient parsing to stay within token constraints

Even now, processing speed is constrained by model limits.

What We Learned

AI pipelines must be fault-tolerant at every stage
Multi-step workflows require careful state management
Designing for a specific user (law students) leads to better, more focused features