inVision Selector — University Admissions Edition

⚠️ SUPPORT TOOL ONLY — This system assists admissions committee decision-making. All final decisions must be made by the committee.

What It Does

inVision Selector is an AI-powered student application evaluation support system for university admissions. It helps admissions committees make more informed decisions by:

• Importing student data from any format — PDFs, DOCX, XLSX, images, audio, video, ZIP archives
• Analyzing applications across 6 dimensions: Academic Merit, Motivation, Extracurriculars, Growth Trajectory, Essay Authenticity, and Recommendations
• Detecting AI-generated essays using stylometric analysis and LLM evaluation
• Scoring candidates on a 0-100 scale with full explainability
• Identifying strengths and weaknesses for each applicant
• Mitigating bias by explicitly checking for demographic/socio-economic proxy scoring
• Managing committee decisions (admit, waitlist, deny, defer, conditional)

Key Features

📁 Universal File Import

Upload any file type for student applications:

• Documents: PDF, DOCX, DOC, TXT, RTF, ODT, TEX
• Spreadsheets: XLSX, XLS, CSV, ODS
• Images: JPG, PNG, GIF, BMP, TIFF, WEBP, HEIC
• Presentations: PPTX, PPT, ODP
• Audio/Video: MP3, WAV, MP4, MOV, etc. (Whisper transcription)
• Archives: ZIP, RAR, 7Z (bundled submissions)

Text is automatically extracted from supported formats. Missing data displays as N/A — never penalized.

⚙️ Configurable LLM Model

Change the AI model from the UI settings panel. Defaults to Qwen 2.5 3B (fast, efficient):

• Qwen 2.5 3B (recommended default)
• Phi-3 Mini, Llama 3.2 3B, Gemma 2 2B
• Mistral 7B, Llama 3.1 8B, Qwen 2.5 7B
• Custom model names

📊 Six Evaluation Dimensions

Dimension	Weight	What It Measures
Academic Merit	25%	GPA, test scores, course rigor, achievements
Motivation	20%	Program fit, genuine interest, career vision
Extracurriculars	20%	Activities, leadership, community involvement
Growth Trajectory	15%	Path traveled, resilience, learning from setbacks
Essay Authenticity	10%	Human-written vs. AI-generated detection
Recommendations	10%	Quality and specificity of recommendation letters

Quick Start

Prerequisites

1. Python 3.10+
2. Ollama installed and running

   # Install Ollama from https://ollama.ai
   # Pull a small model (default):
   ollama pull qwen2.5:3b

Installation & Run

cd invision-selector
pip install -r requirements.txt

# Optional: install document extraction support
pip install PyPDF2 python-docx openpyxl

# Start the server
uvicorn src.main:app --reload --host 0.0.0.0 --port 8000

Windows: Run run.bat

Open http://localhost:8000 — 6 demo students auto-generated on first run.

API Endpoints

Endpoint	Method	Description
/	GET	Dashboard
/api/health	GET	Health check + model info
/api/students	GET/POST	List / create student applications
/api/students/{id}	GET/PUT	Get / update student application
/api/students/{id}/documents	POST	Upload a document
/api/students/{id}/documents/batch	POST	Upload multiple documents
/api/documents/{id}/download	GET	Download a document
/api/students/{id}/analyze	POST	Trigger AI analysis
/api/students/{id}/notes	PUT	Update committee notes
/api/students/{id}/decision	PUT	Record committee decision
/api/ranking	GET	Ranked student list
/api/settings/model	GET/PUT	Get / change LLM model
/api/demo/reset	POST	Reset demo data

Dependencies

• fastapi — Web framework
• uvicorn — ASGI server
• aiosqlite — Async SQLite
• httpx — Async HTTP client (for Ollama)
• openai — Whisper API (optional)
• pydantic — Data validation

Optional (for document extraction)

• PyPDF2 or pdfplumber — PDF text extraction
• python-docx — DOCX text extraction
• openpyxl — XLSX text extraction
• python-pptx — PPTX text extraction

License

Hackathon prototype — for demonstration purposes.