AI‑Powered Physics, Statics & Dynamics Answer Evaluator
SciGradex is a web-based application for teachers to evaluate physics/statics/dynamics answers using AI. It doesn’t just give scores – it provides feedback and suggests correct solutions, making checking more effective.
- Base Model: LLaMA 3 (accessed via Groq API)
- Type: Large Language Model (LLM) optimized for reasoning and context‑aware responses.
- High Reasoning Ability – Performs well on problem‑solving and evaluation tasks.
- Structured Output – Generates scored assessments & feedback in a consistent, easy‑to‑read format.
- Math & Symbol Support – Handles LaTeX‑style math, physics symbols, and technical content effectively.
- Prompt Adaptability – Easily guided using pre‑defined prompt templates for specialized grading tasks.
- Efficient & Fast – Leveraging Groq’s low‑latency infrastructure for near real‑time evaluation.
- Available sizes: 8B & 70B parameters (we used Groq‑hosted 8B for faster inference).
- Weights: Open‑weight (by Meta), but API access via Groq is paid (with a free trial tier).
- Ultra‑low latency inference – ideal for real‑time evaluation.
- Scalable & reliable API for consistent model responses.
- AI-Powered Evaluation – Get a score (0–10), feedback, and a correct answer suggestion.
- Math-Aware – Supports complex formulas, summations (Σ), fractions, symbols (λ, μ, etc.) with MathJax rendering.
- Evaluation History – Search & filter past evaluations.
- Secure Authentication – User registration & login with hashed passwords.
- Modern UI – Clean, responsive design with color-coded scores & interactive controls.
chunks.pkl– Pickled (binary) version of pre-processed prompt chunks for faster loading in production.chunks.json– Stores prompt templates used to communicate with Groq API for consistent evaluation.index.fiass– Vector index (FAISS) for storing & retrieving embeddings, allowing similarity-based search for past questions or prompts.preprocess.py– Cleans and normalizes raw input (removes extra spaces, converts special characters, prepares equations for MathJax rendering).app.py– Flask backend, handles routes (/submit-question,/get-history), connects with the database & AI engine.- scripts/preprocess.py
Step-by-Step:
- Input: Teacher submits a question & student answer (manually).
- Processing:
- Text is fed into Groq API with a carefully designed prompt.
- AI Evaluation:
- Model generates score, feedback, and correct solution suggestion.
- Database Storage: Results are stored in
queriestable for later retrieval. - Frontend Display: Results are rendered with MathJax for proper equation formatting.
- Frontend: HTML5, CSS3, JavaScript, Font Awesome, MathJax
- Backend: Python (Flask)
- Database: SQLite3
- AI Engine: Groq API
(Store these inside an assets/ folder)
(Click to view full workflow: Login → Evaluate Answers → View History)
- Frontend: HTML5, CSS3, JavaScript, Font Awesome, MathJax
- Backend: Python (Flask)
- Database: SQLite3
- AI Engine: Groq API
- Clone the repository:
git clone https://github.com/Anooshakhalid/DeepSeek-Learning-Model--Project.git cd DeepSeek-Learning-Model--Project
This project is licensed under the MIT License – feel free to use, modify, and distribute with attribution.
- Email - anooshakhalid999@gmail.com
- NED University of Engineering & Tech