Hire Me
0
×
AI Portfolio Project Features
- Hybrid RAG Search: Semantic embeddings with FAISS and BM25 fallback
- OpenAI GPT-4o Integration: Primary LLM with quota handling
- Smart Context Management: Maintains 5-message history with multilingual support
- Dynamic Updates: Notion content sync with real-time embedding and index refresh
- Robust Tech Stack: Flask backend, Python AI libraries, responsive frontend
- Security & Performance: Admin authentication, error handling, scalable design
- Deployment Ready: Cross-platform and production-ready architecture
Tech Stack
- Backend: Flask, OpenAI GPT-4o, FAISS, BM25Okapi, Python
- Data: Notion API, JSON, SQLite
- Frontend: HTML5, CSS3, JavaScript, Font Awesome
- AI Libraries: NumPy, scikit-learn, deep-translator, rank-bm25
- Dev & Deployment: Git, Python subprocess, environment variables
- APIs: OpenAI, Notion, RESTful endpoints
Links
×
TalentSynth-Resume Analyzer
Developed an AI-powered resume parser using LlamaParse and Google Gemini 2.5 Flash to extract and semantically analyze candidate data against job descriptions.
Key Features
- Resume Parsing: Advanced document parsing supporting PDFs and DOCX, extracting text and metadata with semantic understanding via Gemini AI.
- Experience & Skill Detection: Automated calculation of candidate experience and identification of technical and soft skills using AI-powered semantic analysis.
- AI-Powered Job Matching: Matches resumes to job descriptions with scoring, strengths and weaknesses analysis, and actionable recommendations.
- Secure User Management: Google OAuth login/signup with Django Allauth, supporting user profiles and real-time analysis results.
- Scalable Full-Stack App: Built with Django and Gunicorn, deployed on Railway with real-time updates and analytics dashboard.
Tech Stack
Django, Gunicorn, llama-parse, llama-index, SQLite
Links
×
Pneumonia Detection using AI
Developed an ensemble deep learning system combining Simple CNN, MobileNetV2, and EfficientNetB0 to classify chest X-ray images into NORMAL and PNEUMONIA. Deployed as a Streamlit web application with Grad-CAM visualizations for model explainability.
Training Strategy
- Simple CNN: Achieved ~94.2% accuracy with AUC ~0.9848 after tuning.
- MobileNetV2: Reached ~95.2% accuracy, AUC ~0.9925 with transfer learning.
- EfficientNetB0: Achieved 94.7% accuracy, AUC ~0.9874 using two-phase training.
- Ensemble: Weighted averaging of predictions improved confidence.
Model Performance
- Accuracy: 96%
- AUC: 0.9941
- Precision: NORMAL 96%, PNEUMONIA 96%
- Recall: NORMAL 95%, PNEUMONIA 97%
- F1 Score: NORMAL 95.5%, PNEUMONIA 96.5%
Tech Stack
Python, TensorFlow/Keras, Streamlit, OpenCV, Grad-CAM, NumPy
Links
×
Linear Algebra Visual Toolkit
Built a web-based toolkit using Python and Streamlit to visualize and solve linear algebra problems. Designed to help users bridge theoretical concepts with practical applications.
Key Features
- Gaussian Elimination solver with step-by-step visualization
- 2D/3D matrix transformation visualizer
- PCA implementation from scratch using eigen decomposition
- Real-time educational tool for exploring concepts interactively
Tech Stack
Python, Streamlit, NumPy, Plotly, Matplotlib