AI File Search Assistant [WIP]

A desktop application for semantic, AI-powered local file search — built entirely in Python, operating fully offline.

Project Synopsis

Abstract

AI File Search Assistant is a desktop application that enables users to search files on their local computer using natural language queries. Instead of remembering exact file names, users can type queries such as "my Python notes" or "invoice from January", and the system retrieves the most relevant files based on their content and meaning.

The application scans selected folders, extracts text from supported file formats, generates semantic embeddings using Sentence Transformers, and stores metadata in SQLite. FAISS is used for fast vector similarity search. The system is built entirely in Python with a PySide6 graphical interface and operates fully offline, ensuring privacy and security.

Objectives

Build a desktop application for semantic file search
Extract content from multiple file types
Generate embeddings using AI models
Store file metadata in SQLite
Use FAISS for fast similarity search
Provide a user-friendly GUI
Ensure complete offline operation

Scope

The system supports the following file types:

Category	Formats
Documents	`.pdf`, `.docx`, `.txt`
Data	`.csv`
Source Code	`.py`, `.java`, `.c`, `.cpp`, `.js`

Hardware & Software Configuration

Hardware Requirements

Component	Minimum Requirement
Processor	Intel Core i5 / AMD Ryzen 5
RAM	8 GB
Storage	256 GB SSD
GPU	Optional
Internet	Required only for installation

Software Requirements

Component	Technology
Programming Language	Python 3.10+
GUI Framework	PySide6
Database	SQLite
Embedding Model	Sentence Transformers (`all-MiniLM-L6-v2`)
Vector Search	FAISS
PDF Parsing	PyMuPDF
DOCX Parsing	python-docx
IDE	VS Code / PyCharm
Version Control	Git & GitHub
Operating System	Windows 10/11 or Linux

Module Description

The system is divided into 10 modules, each handling a distinct responsibility.

1. User Management Module

Handles user registration, login authentication, search history tracking, and user settings.

2. Folder Management Module

Handles selecting folders to index, storing indexed folder paths, and enabling or disabling folders.

3. File Indexing Module

Handles recursive folder scanning, metadata collection, and detecting modified or new files.

4. Content Extraction Module

Extracts text from the following formats: PDF, DOCX, TXT, CSV, and source code files.

5. Embedding Generation Module

Converts extracted text into semantic vectors using the Sentence Transformers library (all-MiniLM-L6-v2).

6. Vector Index Module

Stores embedding vectors in FAISS and performs nearest-neighbour similarity search.

7. Semantic Search Module

Processes natural language queries, generates a query embedding, and retrieves the most relevant files.

8. Database Management Module

Manages all SQLite database operations — inserts, updates, queries, and schema management.

9. File Preview Module

Displays extracted text content for a selected search result inside the GUI.

10. Desktop GUI Module

Provides the full graphical interface including: login form, main dashboard, search interface, and preview pane.

Data Flow Diagrams

1. User Management Module

flowchart TD
    U[User] -->|Register / Login| UM[User Management Module]
    UM -->|Validate Credentials| DB[(USER Table)]
    DB -->|User Data| UM
    UM -->|Login Success / Failure| U

2. Folder Management Module

flowchart TD
    U[User] -->|Select Folder| FM[Folder Management Module]
    FM -->|Save Folder Details| DB[(FOLDER Table)]
    FM -->|Start Indexing Request| FI[File Indexing Module]

3. File Indexing Module

flowchart TD
    F[Selected Folder] --> FI[File Indexing Module]
    FI -->|Read Files & Metadata| FS[File System]
    FI -->|Store Metadata| DB[(FILE Table)]
    FI -->|Send File Paths| CE[Content Extraction Module]

4. Content Extraction Module

flowchart TD
    FI[File Indexing Module] -->|File Path| CE[Content Extraction Module]
    CE -->|Extract Text| TXT[Extracted Text]
    TXT --> EG[Embedding Generation Module]
    CE -->|Update Extracted Text| DB[(FILE Table)]

5. Embedding Generation Module

flowchart TD
    TXT[Extracted Text] --> EG[Embedding Generation Module]
    EG -->|Generate Embedding Vector| ST[Sentence Transformer Model]
    EG -->|Store Vector ID| DB[(EMBEDDING Table)]
    EG -->|Add Vector| FAISS[(FAISS Index)]

6. Semantic Search Module

flowchart TD
    U[User] -->|Natural Language Query| SS[Semantic Search Module]
    SS -->|Generate Query Embedding| ST[Sentence Transformer Model]
    SS -->|Similarity Search| FAISS[(FAISS Index)]
    FAISS -->|Top Matching Vector IDs| SS
    SS -->|Fetch File Metadata| DB[(FILE Table)]
    SS -->|Ranked Results| U

7. File Preview Module

flowchart TD
    U[User] -->|Select Search Result| FP[File Preview Module]
    FP -->|Read Extracted Text| DB[(FILE Table)]
    FP -->|Display Preview| U

8. Search History Module

flowchart TD
    SS[Semantic Search Module] -->|Save Query Details| SH[Search History Module]
    SH -->|Insert Search Record| DB[(SEARCH_HISTORY Table)]
    SH -->|Store Result Mapping| DB2[(SEARCH_RESULT Table)]

9. Desktop GUI Module

flowchart TD
    U[User] --> GUI[PySide6 GUI]
    GUI --> UM[User Management Module]
    GUI --> FM[Folder Management Module]
    GUI --> SS[Semantic Search Module]
    GUI --> FP[File Preview Module]
    GUI --> U

10. Overall System DFD (Level 1)

flowchart TD
    U[User]

    U --> GUI[Desktop GUI]

    GUI --> UM[User Management]
    GUI --> FM[Folder Management]
    GUI --> SS[Semantic Search]
    GUI --> FP[File Preview]

    FM --> FI[File Indexing]
    FI --> CE[Content Extraction]
    CE --> EG[Embedding Generation]

    EG --> FAISS[(FAISS Index)]
    EG --> DB[(SQLite Database)]

    SS --> FAISS
    SS --> DB

    FP --> DB
    UM --> DB

Database Design

Entity-Relationship Diagram

erDiagram

    USER {
        int user_id PK
        string username
        string email
        string password_hash
        datetime created_at
    }

    FOLDER {
        int folder_id PK
        int user_id FK
        string folder_name
        string folder_path
        boolean is_active
        datetime added_at
        datetime last_scanned_at
    }

    FILE {
        int file_id PK
        int folder_id FK
        string file_name
        string file_path
        string file_extension
        string mime_type
        int file_size
        datetime created_at
        datetime modified_at
        datetime indexed_at
        string content_hash
        text extracted_text
        boolean is_deleted
    }

    EMBEDDING {
        int embedding_id PK
        int file_id FK
        int faiss_vector_id
        string model_name
        int vector_dimension
        datetime created_at
    }

    SEARCH_HISTORY {
        int search_id PK
        int user_id FK
        string query_text
        int results_count
        datetime searched_at
    }

    SEARCH_RESULT {
        int result_id PK
        int search_id FK
        int file_id FK
        float similarity_score
        int rank_position
    }

    INDEXING_LOG {
        int log_id PK
        int file_id FK
        string status
        string error_message
        float processing_time
        datetime indexed_at
    }

    USER ||--o{ FOLDER : owns
    USER ||--o{ SEARCH_HISTORY : performs
    FOLDER ||--o{ FILE : contains
    FILE ||--|| EMBEDDING : has
    FILE ||--o{ INDEXING_LOG : generates
    FILE ||--o{ SEARCH_RESULT : appears_in
    SEARCH_HISTORY ||--o{ SEARCH_RESULT : returns

Database Connectivity

The project uses SQLite — a serverless, lightweight, and fast database engine well-suited for desktop applications.

import sqlite3

def connect():
    return sqlite3.connect("data/file_index.db")

Why SQLite?

Serverless and zero-configuration
Lightweight with a small footprint
Fast for read-heavy workloads
Easy to distribute with the application
Ideal for single-user desktop applications

Table Schemas

USER Table

Column	Type	Constraint
user_id	INTEGER	PRIMARY KEY
username	TEXT	UNIQUE
email	TEXT	UNIQUE
password_hash	TEXT	NOT NULL
created_at	DATETIME	DEFAULT CURRENT_TIMESTAMP

FOLDER Table

Column	Type	Constraint
folder_id	INTEGER	PRIMARY KEY
user_id	INTEGER	FOREIGN KEY
folder_path	TEXT	UNIQUE
folder_name	TEXT	NOT NULL
is_active	BOOLEAN	DEFAULT 1
added_at	DATETIME	DEFAULT CURRENT_TIMESTAMP

FILE Table

Column	Type	Constraint
file_id	INTEGER	PRIMARY KEY
folder_id	INTEGER	FOREIGN KEY
file_name	TEXT	NOT NULL
file_path	TEXT	UNIQUE
file_extension	TEXT	NOT NULL
mime_type	TEXT	—
file_size	INTEGER	—
modified_at	DATETIME	—
indexed_at	DATETIME	—
content_hash	TEXT	—
extracted_text	TEXT	—

EMBEDDING Table

Column	Type	Constraint
embedding_id	INTEGER	PRIMARY KEY
file_id	INTEGER	FOREIGN KEY
faiss_vector_id	INTEGER	UNIQUE
model_name	TEXT	—
vector_dimension	INTEGER	—
created_at	DATETIME	DEFAULT CURRENT_TIMESTAMP

SEARCH_HISTORY Table

Column	Type	Constraint
search_id	INTEGER	PRIMARY KEY
user_id	INTEGER	FOREIGN KEY
query_text	TEXT	NOT NULL
results_count	INTEGER	—
searched_at	DATETIME	DEFAULT CURRENT_TIMESTAMP

SEARCH_RESULT Table

Column	Type	Constraint
result_id	INTEGER	PRIMARY KEY
search_id	INTEGER	FOREIGN KEY
file_id	INTEGER	FOREIGN KEY
similarity_score	REAL	—
rank_position	INTEGER	—

INDEXING_LOG Table

Column	Type	Constraint
log_id	INTEGER	PRIMARY KEY
file_id	INTEGER	FOREIGN KEY
status	TEXT	—
error_message	TEXT	—
processing_time	REAL	—
indexed_at	DATETIME	DEFAULT CURRENT_TIMESTAMP

Form Design

Element	Type
Username	Text Input
Password	Password Input
Login	Button
Register	Button

Registration Form

Element	Type
Username	Text Input
Email	Text Input
Password	Password Input
Confirm Password	Password Input
Register	Button
Back to Login	Button

Main Dashboard

The dashboard is divided into four sections:

Search Section

Search bar
Search button

Folder Section

Add Folder
Remove Folder
Reindex button

Filters

File type filter
Date range filter
Size range filter

Results Table

Column	Description
File Name	Name of the matched file
Type	File extension/type
Similarity Score	Relevance score from FAISS
Modified Date	Last modified timestamp

Preview Pane — displays extracted text content of the selected result.

Settings Form

Setting	Options
Model Selection	Choose embedding model
Theme	Light / Dark
Max Search Results	Configurable integer

Conclusion

The AI File Search Assistant successfully demonstrates how artificial intelligence can be applied to improve file retrieval on personal computers. By combining content extraction, semantic embeddings, FAISS vector search, SQLite database management, and a PySide6 graphical interface, the system allows users to locate files based on meaning rather than exact file names.

The application operates fully offline, ensuring complete privacy and making it suitable for personal and professional use. The project showcases practical integration of NLP, machine learning, databases, and desktop application development, providing a strong foundation for future enhancements such as:

Voice-based search
Image content search
Cross-platform packaging (macOS, Linux)
Cloud synchronisation support

AI File Search Assistant [WIP]

Project Synopsis

Abstract

Objectives

Scope

Hardware & Software Configuration

Hardware Requirements

Software Requirements

Module Description

1. User Management Module

2. Folder Management Module

3. File Indexing Module

4. Content Extraction Module

5. Embedding Generation Module

6. Vector Index Module

7. Semantic Search Module

8. Database Management Module

9. File Preview Module

10. Desktop GUI Module

Data Flow Diagrams

1. User Management Module

2. Folder Management Module

3. File Indexing Module

4. Content Extraction Module

5. Embedding Generation Module

6. Semantic Search Module

7. File Preview Module

8. Search History Module

9. Desktop GUI Module

10. Overall System DFD (Level 1)

Database Design

Entity-Relationship Diagram

Database Connectivity

Table Schemas

USER Table

FOLDER Table

FILE Table

EMBEDDING Table

SEARCH_HISTORY Table

SEARCH_RESULT Table

INDEXING_LOG Table

Form Design

Login Form

Registration Form

Main Dashboard

Settings Form

Conclusion