• AI Librarian

  • AI Librarian

  • AI Librarian

  • AI Librarian

  • AI Librarian

  • AI Librarian

  • AI Librarian

  • AI Librarian

  • AI Librarian

Product Manager Accelerator - Remote

Product Manager Accelerator - Remote

Overview

Overview

Overview

Small and mid-size startups across industries struggle with chaotic, siloed file systems—everything from design files and spreadsheets to PDFs, audio clips, and emails. Teams waste hours hunting for assets, duplicating work, and losing context amid inconsistent naming and folder structures. Our AI-Librarian consolidates every file type, auto-extracts metadata, enforces conventions, and builds semantic relationships so anyone can get instant, accurate results. By removing the friction of file management, we free teams to focus on creativity and core business goals.

Small and mid-size startups across industries struggle with chaotic, siloed file systems—everything from design files and spreadsheets to PDFs, audio clips, and emails. Teams waste hours hunting for assets, duplicating work, and losing context amid inconsistent naming and folder structures. Our AI-Librarian consolidates every file type, auto-extracts metadata, enforces conventions, and builds semantic relationships so anyone can get instant, accurate results. By removing the friction of file management, we free teams to focus on creativity and core business goals.

Tools

Tools

Tools

Figma

Figma

Illustrator

Illustrator

Photoshop

Photoshop

Devin AI

Devin AI

Timeline

Timeline

Timeline

Start: Feb 2025

Start: Feb 2025

End: Apr 2025

End: Apr 2025

Problem Statement

Problem Statement

Problem Statement

Small and mid‑size companies generate a torrent of files—spreadsheets, design files, slide decks, Word docs, PDFs, audio notes and email threads—scattered across local drives, cloud services and messaging apps. Inconsistent naming and disorganized folders turn search into a guessing game, while version confusion stalls decisions, duplicates work and erodes team trust. Critical annotations get buried, contractual documents go missing and compliance risks rise, all dragging productivity and morale down. The AI‑Powered Document Librarian remedies this by ingesting every file type, using AI to extract context and metadata, and linking related resources so that natural‑language queries deliver precise results, instantly restoring teams’ focus on their core work

My Role

My Role

My Role

As Design Lead, I defined the information architecture by running collaborative workshops, sketched low-fidelity wireframes, then translated those into interactive Figma prototypes and polished visual designs. To ensure consistency and scalability, I created a component library and documented usage guidelines. Each week I met with frontend developers to align on the React implementation, review edge cases and refine interactions and accessibility details. This continuous collaboration guaranteed that the finished product matched our design vision down to the smallest hover state or transition.

As Design Lead, I defined the information architecture by running collaborative workshops, sketched low-fidelity wireframes, then translated those into interactive Figma prototypes and polished visual designs. To ensure consistency and scalability, I created a component library and documented usage guidelines. Each week I met with frontend developers to align on the React implementation, review edge cases and refine interactions and accessibility details. This continuous collaboration guaranteed that the finished product matched our design vision down to the smallest hover state or transition.

Current Status

Current Status

Current Status

Research Insights & Key Pain Points
We spoke with eight professionals at startups and mid-size companies across marketing, operations, design and finance to understand their document workflows. Four core pain points emerged:

  • Disparate File Formats (CAD drawings, PPT decks, Excel exports, JPEGs, MP3s, DOCX files, email threads) leading to constant context switching.

  • Inconsistent Naming & Folder Structures causing duplicate assets and wasted search time when not sure of the naming.

  • Lost Collaboration Threads as feedback lives in Slack, email or comment tools rather than attached to the document itself.

  • Lack of Historical Insights forcing teams to reinvent past solutions and missing out on lessons learned.

Research Insights & Key Pain Points
We spoke with eight professionals at startups and mid-size companies across marketing, operations, design and finance to understand their document workflows. Four core pain points emerged:

  • Disparate File Formats (CAD drawings, PPT decks, Excel exports, JPEGs, MP3s, DOCX files, email threads) leading to constant context switching.

  • Inconsistent Naming & Folder Structures causing duplicate assets and wasted search time when not sure of the naming.

  • Lost Collaboration Threads as feedback lives in Slack, email or comment tools rather than attached to the document itself.

  • Lack of Historical Insights forcing teams to reinvent past solutions and missing out on lessons learned.

Feature Breakdown & MVP Boundaries

The MVP delivers a streamlined upload experience, powerful retrieval via an AI chatbot, flexible search, relationship visualization, and standard document viewing and sorting. Advanced analytics and external integrations are deferred to later releases.

  • Unified Upload & Metadata Extraction
    Users drag‑and‑drop any file type—PDF, CAD, PPT, DOCX, JPEG, MP3, email thread, etc.—and the system immediately performs OCR, semantic parsing, and schema‑driven tagging (project, client, department, date, keywords) as part of the same workflow. Progress indicators show real‑time status and any auto‑tagging suggestions can be reviewed before finalizing.

  • AI Chatbot Retrieval (RAG‑Powered)
    Leveraging a Retrieval‑Augmented Generation framework, the chatbot is the primary interface for file retrieval. Users simply ask in natural language (“Show me the latest vendor invoices for Project Orion” or “Find design mockups that reference our new color palette”) and the system fetches, synthesizes, and ranks relevant documents, even suggesting related files based on semantic context.

  • Faceted & Full‑Text Search
    For users who prefer traditional querying, the library supports combined filters—file type, upload date, tags—and full‑text search across all ingested content. Results appear with inline previews and can be sorted by relevance, date, or custom fields.

  • Relationship Graph Visualization
    An interactive node‑and‑edge map reveals connections among files, projects, versions, and collaborators. Clicking any node brings up the document viewer, allowing users to navigate complex document networks with ease.

  • Document Viewing & Sorting
    Within the library, each document opens in a built‑in viewer supporting pagination, zoom, and annotations. Users can sort lists by name, date, file type, or metadata fields, ensuring quick access without leaving the interface.

Information Architecture & User Flow

Information Architecture & User Flow

Information Architecture & User Flow

Homepage branches into four sections:

  1. Upload → Drag‑drop any file → “Edit & Confirm Metadata” → Document Info

  2. Explore (Linking) → Multidimensional graph with filter controls → Click node → Document Info

  3. Documents → Full‑text search + metadata filters → Pick result → Document Info

  4. Retrieval / AI Chat → RAG‑powered chat with in‑chat filters and prompt lists → Select file → Document Info

Key User Flows

  1. Upload Flow: Homepage → Upload → Confirm metadata → Document Info

  2. Linking Flow: Homepage → Explore → Adjust filters → Click node → Document Info

  3. Search Flow: Homepage → Documents → Search or filter → Document Info

  4. Chat Flow: Homepage → AI Chat → Natural‑language query → Refine via filters/prompts → Document Info

Prototyping & Validation

Low-Fidelity: Greyscale sketches and wireframes tested with 5 users to validate user interactions and flow usability.

High-Fidelity: Clickable Figma and webpage prototype; weekly demos with engineers and PMs refined micro-interactions (hover states, loading indicators, error handling).

Visual Refinement & Final Deliverables

Visual Refinement & Final Deliverables

Visual Refinement & Final Deliverables

  • Interactivity: Hover states, transition animations on graph zoom

  • Accessibility: WCAG-compliant color contrast, keyboard navigation

  • Final Assets: Clickable Figma prototype link

Impact & Next Steps

Impact & Next Steps

Impact & Next Steps

In early trials, the AI‑Librarian cut file‑retrieval time by roughly 70%, significantly boosting cross‑team efficiency. The system remains in active development and is slated for launch shortly. Upcoming enhancements will include robust admin controls for team management and permissions, a more refined RAG‑based AI framework to improve the precision and relevance of conversational retrieval, user‑driven naming‑template support, in‑app onboarding tutorials, and expanded connectors to common storage platforms.

©2025 Boris Hu. All Rights Reserved.

©2025 Boris Hu. All Rights Reserved.

©2025 Boris Hu. All Rights Reserved.