Anurag Tiwari

GitWise AI — GitHub Repository Q&A System

Built a Retrieval-Augmented Generation (RAG) system to answer natural language questions on GitHub repositories.

• Core System: Designed pipeline for repository ingestion, code chunking, and embedding generation.
• Retrieval: Implemented hybrid search using MiniLM embeddings with BGE reranker for high-quality results.
• LLM Integration: Used Groq LLM to generate context-aware answers from retrieved code snippets.
• Database: Used Qdrant vector database for efficient semantic search with metadata filtering.
• Architecture: Built modular system with separate ingestion, retrieval, and response layers.
• Evaluation: Created evaluation pipeline to measure retrieval accuracy and response quality.
• Tech Stack: Python, FastAPI, LangChain, SentenceTransformers, Qdrant, Groq LLM, Streamlit, Docker

Power Consumption Prediction

Predicts household electricity consumption using historical sensor data and time-based features.

• Performance: MAE: ~0.0177 | RMSE: ~0.0329 | R²: ~0.9987
• Features: Historical sensor readings (Global reactive power, Voltage, Current, Sub-metering 1-3), time features (hour, day of week, month), and lag features to capture temporal dependencies.
• Tech Stack: Python, Pandas, NumPy, scikit-learn, Streamlit, Docker
• Highlights:
  • Preprocessing pipeline handles missing values and ensures numeric types for all features.
  • Lag features allow the model to capture short-term dependencies in power consumption.
  • Random Forest Regressor with feature selection (SelectKBest) to focus on the most important predictors and reduce overfitting.
  • Interactive Streamlit UI supports both single and batch predictions, with metrics displayed when actual values are available.

Telco Customer Churn Prediction

Predicts whether a telecom customer is likely to churn based on demographic, account, and service usage features.

• Performance: Accuracy: 82.1% | ROC-AUC: 0.87
• Features: Tenure, MonthlyCharges, TotalCharges, gender, SeniorCitizen, Partner, Dependents, PhoneService, MultipleLines, InternetService, OnlineSecurity, OnlineBackup, DeviceProtection, TechSupport, StreamingTV, StreamingMovies, Contract, PaperlessBilling, PaymentMethod
• Tech Stack: Python, Pandas, NumPy, scikit-learn, Streamlit, Docker
• Highlights:
  • Preprocessing pipeline handles missing values, scales numeric features, and one-hot encodes categorical features.
  • Random Forest classifier with class balancing, controlled tree depth, and optimized splits to reduce overfitting.
  • Interactive Streamlit UI supports single and batch predictions with churn probability and churn rate metrics.
  • Correlation analysis of numeric features to improve model robustness.

Titanic Survival Predictor

Predicts Titanic passenger survival using demographic and travel information. Supports single predictions via UI and batch predictions via CSV upload.

• Performance: Accuracy: 82.3% | ROC-AUC: 0.87 | Cross-Validation: 81.5% ± 2.1%
• Features: Age, Fare, Pclass, Sex, Embarked, Siblings/Spouses, Parents/Children, Title, FamilySize, IsAlone
• Tech Stack: Python, Pandas, NumPy, scikit-learn, Streamlit, Docker
• Highlights:
  • Interactive Streamlit app for single predictions with survival probability.
  • Batch prediction support for CSV files, generating predictions and probabilities for multiple passengers.
  • Custom preprocessing pipeline handling missing values, feature engineering, skewness correction, scaling, and one-hot encoding.
  • Decision Tree classifier with controlled depth and leaf size to reduce overfitting.

AuthVerse — Three Level Authentication System

Built AuthVerse, a secure multi-level authentication system with three layers:

• Text Password: Standard password login.
• Color Password: Users select a combination of colors as a password.
• Image Password: Users tap specific areas on a cropped image to log in.

Tech Stack: PHP, MySQL, HTML, CSS, JavaScript

Elearning App

Developed an E-learning app with three modules:

• Admin: Approves user accounts and manages access.
• Teacher: Uploads lectures and notes for students.
• Student: Accesses learning materials and connects with teachers.

Tech Stack: Java, Android Studio, Retrofit, Firebase, Glide, PHP, MySQL