TaskPilot

• 

  TaskPilot Dashboard — Multi-format document task extraction with priority routing and status management.

Inspiration

Documents, emails, and guides are packed with critical tasks, deadlines, and actionable takeaways, but manually reading through lengthy files to identify work items is time-consuming and error-prone. We built TaskPilot to automate document triage—transforming static PDFs, plain text files, and saved email (.eml) messages into structured, prioritized task lists instantly using cutting-edge AI inference.

What it does

TaskPilot is an intelligent AI document analyzer that ingests user files and automatically extracts actionable tasks:

• Multi-Format Ingestion: Accepts PDF documents, plain text files (.txt), and raw email formats (.eml) including attached content.
• Instant Summarization: Generates a high-level summary of the entire document.
• Structured Task Extraction: Parses out actionable task items complete with titles, deadlines (YYYY-MM-DD), priority levels (High, Medium, Low), and detailed action requirements.
• Database & Task Management: Persists extracted data locally in SQLite, supporting task status toggles (Pending / Completed) and 1-click .ics calendar export.
• Future Action Items Sidebar: Automatically routes low-priority or long-term action items into a dedicated side panel.

How we built it

• AI Model: Powered by NVIDIA Nemotron-3.5-Lightning hosted on Nebius Token Factory API.
• Backend: Built with FastAPI (Python 3.14) and SQLite / SQLAlchemy to manage database persistence, file uploads, document text parsing using pypdf and Python's built-in email parser, and structured JSON extraction.
• Frontend: Designed with Next.js, React, and Tailwind CSS for a responsive dark-mode user experience.

Challenges we faced

1. Schema Consistency: Ensuring LLM outputs consistently matched schema requirements without extra markdown formatting. We solved this by refining system prompts and creating structured parsing handling in FastAPI.
2. Multi-Format Parsing: Handling binary raw stream differences between .pdf, .txt, and .eml email structures while extracting clean text strings for token efficiency.
3. API Integration: Setting up low-latency inference routing through Nebius Cloud to leverage NVIDIA GPU acceleration seamlessly.

Accomplishments that we're proud of

• Achieved fast sub-second task extraction from complex multipage PDF documents and emails.
• Built a clean, modern UI with status management, calendar sync, and dual-column task categorization.
• Handled edge cases seamlessly to guarantee zero frontend crashes even when document structures vary significantly.

What we learned

• Best practices for prompt engineering to enforce strict JSON output schemas on NVIDIA Nemotron models.
• How to leverage Nebius cloud infrastructure for high-throughput OpenAI-compatible API calls.
• Effective state management patterns for multi-file format parsing pipelines in FastAPI and Next.js.

What's next for TaskPilot

• Google Calendar API Sync: Direct cloud authentication and 1-click Google Calendar push.
• Multi-User Authentication: Cloud hosting with PostgreSQL and OAuth user login.
• Deep Parsing for Zip Attachments: Expanding parser capability to automatically uncompress .zip attachments inside .eml files.

Built With

• fastapi
• nebius
• next.js
• nvidia-nemotron
• pypdf
• python
• react
• sqlite
• tailwind-css