SynthiCal AI PRD / Study — Tharun Poduru

← Back to SynthiCal AI

Product Name: SynthiCal AI
Version: 2.0
Date: April 15th 2025
Product Manager & Developer: Tharun Poduru

1. Introduction

Goals:

To provide users with an effortless, AI-powered method for extracting event details from any unstructured source — text, URLs, documents, images, and audio recordings.
To eliminate manual calendar data entry by transforming raw information into structured, downloadable calendar events.
To offer a privacy-first tool that requires no sign-up, stores no user data, and deletes uploaded files after processing.

Vision: SynthiCal AI aims to become the universal bridge between unstructured event information and structured calendar entries. Whether a user receives a conference schedule in a PDF, a meeting invitation in a screenshot, event details on a web page, or verbal instructions in a voice memo, SynthiCal AI extracts every detail and delivers it directly to their calendar of choice.

2. Target Audience

Primary Users:

Busy Professionals: Receiving event details across email, Slack, documents, and web links — needing a fast way to get them into their calendar.
Students: Managing class schedules, club events, and academic deadlines scattered across syllabi, emails, and campus web pages.
Event Organizers: Processing event listings from partner websites, flyers, and promotional materials.

Secondary Users:

Administrative Assistants: Extracting meeting details from correspondence and documents on behalf of others.
Freelancers & Consultants: Managing client meetings and deadlines received in varied formats.

3. User Stories & Use Cases

US1: As a professional, I want to paste a block of text containing meeting details and receive a structured event I can add to Google Calendar, so I don't have to manually type out dates and times.
US2: As a student, I want to enter a URL for a campus events page and extract all listed events at once, so I can batch-add them to my calendar.
US3: As an event organizer, I want to upload an image of a conference flyer and have the event details extracted automatically, so I can quickly create calendar entries for promotion.
US4: As a user in a meeting, I want to record a voice memo describing upcoming events and have the dates, times, and locations extracted, so I can capture scheduling details hands-free.
US5: As any user, I want to upload a PDF itinerary and extract all events from it, so I don't have to manually enter each item.
US6: As any user, I want to select specific extracted events and download only those as a single .ics file, so I can choose which events to add to my calendar.
US7: As a user in a different timezone, I want to select my timezone before extraction so that relative date references like "tomorrow" or "next Friday" are resolved correctly for my location.

4. Key Features

F1: Multi-Modal Input System

F1.1: Text input via a textarea for pasting or typing event information.
F1.2: URL input that fetches, parses, and extracts events from web pages.
F1.3: Document upload supporting PDF, DOC, DOCX, TXT, and RTF files (max 25 MB).
F1.4: Image upload supporting PNG, JPG, JPEG, WebP, and HEIF formats (max 25 MB).
F1.5: In-browser audio recording with 60-second limit, playback preview, and WebM/MP4 capture.
F1.6: Tabbed interface for switching between input modes (Text, Document, Image, Audio, URL).

F2: AI-Powered Event Extraction

F2.1: Google Gemini 3.1-pro-preview processes all input types for event extraction.
F2.2: Structured JSON output with title, start/end datetime (UTC ISO), location, and description.
F2.3: Timezone-aware date resolution using current UTC time, user timezone, and local time context.
F2.4: Automatic end-time inference (start + 1 hour) when only a start time is provided.
F2.5: Flexible response normalization handling arrays, nested objects, and single events.

F3: Intelligent URL Processing

F3.1: Primary extraction from JSON-LD schema.org/Event structured data embedded in page markup.
F3.2: Fallback to AI extraction using a focused 4,000-character content snippet when no structured data exists.
F3.3: HTML parsing via JSDOM with content targeting (main, article, .content, #content, .post, .entry-content).
F3.4: Graceful error recovery returning a placeholder event rather than a failure state.

F4: Multimodal File Processing

F4.1: Resumable upload to Gemini Files API with proper MIME type handling.
F4.2: Polling mechanism for file activation status (up to 45 seconds).
F4.3: Multimodal content generation combining file reference with extraction prompt.
F4.4: Automatic file deletion from Gemini servers after processing.

F5: Event Display & Management

F5.1: Responsive card grid layout (1 / 2 / 3 columns) displaying extracted events.
F5.2: Individual event selection with visual highlight (ring border).
F5.3: Select-all and multi-select capabilities for batch operations.
F5.4: Formatted date/time display with weekday, date, time, and timezone abbreviation.
F5.5: Description rendering with [br] to <br/> conversion for formatted text.
F5.6: Source link display for URL-extracted events.

F6: Calendar Export

F6.1: "Add to Calendar" button per event with dropdown: Google Calendar, Apple Calendar, Outlook.com, iCal.
F6.2: Individual .ics file download per event card.
F6.3: Download selected events as a single .ics file.
F6.4: Download all events as a single .ics file.
F6.5: Server-side ICS generation using the ics library with UTC date components.

F7: Analytics & Monitoring

F7.1: Google Analytics 4 integration (opt-in via environment variable).
F7.2: Event logging for: extraction_succeeded, extraction_failed, download_all_ics, download_selected_ics.
F7.3: Extraction type and event count tracked per successful extraction.
F7.4: Error category logged for failed extractions.

5. Non-Functional Requirements

NFR1: The application shall support all modern desktop and mobile browsers (Chrome, Firefox, Safari, Edge).
NFR2: Text and URL extraction shall return results within 10 seconds for typical inputs.
NFR3: File processing (upload, analysis, cleanup) shall complete within 30 seconds for files up to 25 MB.
NFR4: The UI shall be responsive and usable on screen widths from 320px to 2560px.
NFR5: No user data shall be stored beyond the browser session lifecycle.
NFR6: Uploaded files shall be deleted from Gemini's servers immediately after processing.
NFR7: The system shall handle API errors gracefully with user-friendly error messages.

6. Key Technical Decisions & Considerations

API Strategy: All AI processing happens server-side via Firebase Cloud Functions (Gen 2). The Gemini API key is stored as a Firebase Secret, never exposed to the client. Text uses the @google/generative-ai SDK; file processing uses the Gemini REST API directly for resumable uploads.
URL Processing Pipeline: A tiered approach (JSON-LD first, AI fallback second) balances speed, accuracy, and cost. Structured data extraction avoids AI API calls entirely for well-marked-up pages.
Multipart Handling: Firebase Functions receive raw body buffers, so Busboy is fed via bb.end(req.rawBody) instead of the standard stream piping approach.
Technology Stack: React 19 with Vite 7 for the frontend; Express 5 on Node.js 20 for the backend; Tailwind CSS 4 for styling; add-to-calendar-button for cross-platform calendar integration.
No Authentication: Deliberately omitted to minimize friction. The tool is designed for quick, anonymous use — paste, extract, export.

7. Success Metrics (KPIs)

User Acquisition: 500+ unique users in the first 3 months
Extraction Accuracy: >85% of event data correctly structured across all input types
Multi-Modal Usage: >20% of extractions use non-text inputs (URL, file, image, audio)
Engagement: Average 2+ events extracted per user session
Export Rate: >60% of extracted events result in a calendar export action
Uptime: 99.5% monthly availability target

8. Future Considerations

Direct Google Calendar API integration with OAuth for one-tap event creation
Chrome extension for contextual extraction from any web page
Email integration for automatic extraction from inbox messages
User accounts with extraction history and saved event templates
Recurring event pattern detection from descriptions
Batch processing for multi-page documents and long PDFs
Team sharing — extract events and share structured results with collaborators

← Back to project overview

← Back to SynthiCal AI

Product Name: SynthiCal AI
Version: 2.0
Date: April 15th 2025
Product Manager & Developer: Tharun Poduru

1. Introduction

Goals:

To provide users with an effortless, AI-powered method for extracting event details from any unstructured source — text, URLs, documents, images, and audio recordings.
To eliminate manual calendar data entry by transforming raw information into structured, downloadable calendar events.
To offer a privacy-first tool that requires no sign-up, stores no user data, and deletes uploaded files after processing.

2. Target Audience

Primary Users:

Busy Professionals: Receiving event details across email, Slack, documents, and web links — needing a fast way to get them into their calendar.
Students: Managing class schedules, club events, and academic deadlines scattered across syllabi, emails, and campus web pages.
Event Organizers: Processing event listings from partner websites, flyers, and promotional materials.

Secondary Users:

Administrative Assistants: Extracting meeting details from correspondence and documents on behalf of others.
Freelancers & Consultants: Managing client meetings and deadlines received in varied formats.

3. User Stories & Use Cases

US1: As a professional, I want to paste a block of text containing meeting details and receive a structured event I can add to Google Calendar, so I don't have to manually type out dates and times.
US2: As a student, I want to enter a URL for a campus events page and extract all listed events at once, so I can batch-add them to my calendar.
US3: As an event organizer, I want to upload an image of a conference flyer and have the event details extracted automatically, so I can quickly create calendar entries for promotion.
US4: As a user in a meeting, I want to record a voice memo describing upcoming events and have the dates, times, and locations extracted, so I can capture scheduling details hands-free.
US5: As any user, I want to upload a PDF itinerary and extract all events from it, so I don't have to manually enter each item.
US6: As any user, I want to select specific extracted events and download only those as a single .ics file, so I can choose which events to add to my calendar.
US7: As a user in a different timezone, I want to select my timezone before extraction so that relative date references like "tomorrow" or "next Friday" are resolved correctly for my location.

4. Key Features

F1: Multi-Modal Input System

F1.1: Text input via a textarea for pasting or typing event information.
F1.2: URL input that fetches, parses, and extracts events from web pages.
F1.3: Document upload supporting PDF, DOC, DOCX, TXT, and RTF files (max 25 MB).
F1.4: Image upload supporting PNG, JPG, JPEG, WebP, and HEIF formats (max 25 MB).
F1.5: In-browser audio recording with 60-second limit, playback preview, and WebM/MP4 capture.
F1.6: Tabbed interface for switching between input modes (Text, Document, Image, Audio, URL).

F2: AI-Powered Event Extraction

F2.1: Google Gemini 3.1-pro-preview processes all input types for event extraction.
F2.2: Structured JSON output with title, start/end datetime (UTC ISO), location, and description.
F2.3: Timezone-aware date resolution using current UTC time, user timezone, and local time context.
F2.4: Automatic end-time inference (start + 1 hour) when only a start time is provided.
F2.5: Flexible response normalization handling arrays, nested objects, and single events.

F3: Intelligent URL Processing

F3.1: Primary extraction from JSON-LD schema.org/Event structured data embedded in page markup.
F3.2: Fallback to AI extraction using a focused 4,000-character content snippet when no structured data exists.
F3.3: HTML parsing via JSDOM with content targeting (main, article, .content, #content, .post, .entry-content).
F3.4: Graceful error recovery returning a placeholder event rather than a failure state.

F4: Multimodal File Processing

F4.1: Resumable upload to Gemini Files API with proper MIME type handling.
F4.2: Polling mechanism for file activation status (up to 45 seconds).
F4.3: Multimodal content generation combining file reference with extraction prompt.
F4.4: Automatic file deletion from Gemini servers after processing.

F5: Event Display & Management

F5.1: Responsive card grid layout (1 / 2 / 3 columns) displaying extracted events.
F5.2: Individual event selection with visual highlight (ring border).
F5.3: Select-all and multi-select capabilities for batch operations.
F5.4: Formatted date/time display with weekday, date, time, and timezone abbreviation.
F5.5: Description rendering with [br] to <br/> conversion for formatted text.
F5.6: Source link display for URL-extracted events.

F6: Calendar Export

F6.1: "Add to Calendar" button per event with dropdown: Google Calendar, Apple Calendar, Outlook.com, iCal.
F6.2: Individual .ics file download per event card.
F6.3: Download selected events as a single .ics file.
F6.4: Download all events as a single .ics file.
F6.5: Server-side ICS generation using the ics library with UTC date components.

F7: Analytics & Monitoring

F7.1: Google Analytics 4 integration (opt-in via environment variable).
F7.2: Event logging for: extraction_succeeded, extraction_failed, download_all_ics, download_selected_ics.
F7.3: Extraction type and event count tracked per successful extraction.
F7.4: Error category logged for failed extractions.

5. Non-Functional Requirements

NFR1: The application shall support all modern desktop and mobile browsers (Chrome, Firefox, Safari, Edge).
NFR2: Text and URL extraction shall return results within 10 seconds for typical inputs.
NFR3: File processing (upload, analysis, cleanup) shall complete within 30 seconds for files up to 25 MB.
NFR4: The UI shall be responsive and usable on screen widths from 320px to 2560px.
NFR5: No user data shall be stored beyond the browser session lifecycle.
NFR6: Uploaded files shall be deleted from Gemini's servers immediately after processing.
NFR7: The system shall handle API errors gracefully with user-friendly error messages.

6. Key Technical Decisions & Considerations

API Strategy: All AI processing happens server-side via Firebase Cloud Functions (Gen 2). The Gemini API key is stored as a Firebase Secret, never exposed to the client. Text uses the @google/generative-ai SDK; file processing uses the Gemini REST API directly for resumable uploads.
URL Processing Pipeline: A tiered approach (JSON-LD first, AI fallback second) balances speed, accuracy, and cost. Structured data extraction avoids AI API calls entirely for well-marked-up pages.
Multipart Handling: Firebase Functions receive raw body buffers, so Busboy is fed via bb.end(req.rawBody) instead of the standard stream piping approach.
Technology Stack: React 19 with Vite 7 for the frontend; Express 5 on Node.js 20 for the backend; Tailwind CSS 4 for styling; add-to-calendar-button for cross-platform calendar integration.
No Authentication: Deliberately omitted to minimize friction. The tool is designed for quick, anonymous use — paste, extract, export.

7. Success Metrics (KPIs)

User Acquisition: 500+ unique users in the first 3 months
Extraction Accuracy: >85% of event data correctly structured across all input types
Multi-Modal Usage: >20% of extractions use non-text inputs (URL, file, image, audio)
Engagement: Average 2+ events extracted per user session
Export Rate: >60% of extracted events result in a calendar export action
Uptime: 99.5% monthly availability target

8. Future Considerations

Direct Google Calendar API integration with OAuth for one-tap event creation
Chrome extension for contextual extraction from any web page
Email integration for automatic extraction from inbox messages
User accounts with extraction history and saved event templates
Recurring event pattern detection from descriptions
Batch processing for multi-page documents and long PDFs
Team sharing — extract events and share structured results with collaborators

← Back to project overview

What would you like to know?

Tharun's AI

SynthiCal AI

1. Introduction

2. Target Audience

3. User Stories & Use Cases

4. Key Features

5. Non-Functional Requirements

6. Key Technical Decisions & Considerations

7. Success Metrics (KPIs)

8. Future Considerations

SynthiCal AI

1. Introduction

2. Target Audience

3. User Stories & Use Cases

4. Key Features

5. Non-Functional Requirements

6. Key Technical Decisions & Considerations

7. Success Metrics (KPIs)

8. Future Considerations