XGBoost FastAPI SHAP SMOTE Optuna Imbalanced-learn

Production-grade fraud pipeline with real-time inference & explainability

Precision ████████████████████ 93%
Recall    █████████████████    85%

• 🎯 93% precision / 85% recall on highly imbalanced transaction data
• 🔬 SMOTE oversampling + precision-recall threshold optimization
• ⚡ Low-latency FastAPI inference for live transaction scoring
• 🔍 SHAP values for full model explainability & auditability
• 🔧 Optuna automated hyperparameter search pipeline
• 🖥️ Live prediction visualization UI built-in

NLP XGBoost TF-IDF Flask FinTech Human-in-the-loop

QuickBooks-style ML pipeline for automated financial classification

Accuracy  ██████████████       71.3%
F1 Score  ██████████████       0.71
Best F1   ███████████████      0.75

• 🏦 Classifies ~7,000 financial transactions into 6 categories
• 🧠 TF-IDF + structured features → multi-class XGBoost pipeline
• 🚀 Scalable Flask API with per-prediction confidence scores
• 🔄 Human-in-the-loop feedback loop for continuous improvement
• 🏭 Designed to real FinTech production standards

Bidirectional LSTM Keras NLP Swiggy/Zomato Deep Learning

Stacked deep NLP architecture classifying 465k+ reviews into 5 star ratings

Reviews    ████████████████████  465,000+
Classes    ████████████████████  5 Star Ratings
Model      Stacked BiLSTM + SpatialDropout + L2

• 🗃️ Custom NLP preprocessing pipeline at 465k+ review scale
• 🧠 Stacked Bidirectional LSTM with spatial dropout & L2 reg
• 🎯 Deep semantic context modelling for nuanced sentiment
• ⚖️ Dynamic class weighting to handle review distribution skew
• 🛑 Early Stopping with validation monitoring for generalization

Random Forest Feature Engineering MAE Logistics ML

ML regression model for real-world last-mile delivery time prediction

Algorithm    Random Forest Regressor
Metric       Mean Absolute Error (MAE)
Domain       Logistics & Supply Chain

• 🌲 Random Forest regressor on historical delivery data
• 🔧 Heavy feature engineering: time, distance, order type
• 📉 Optimized for MAE — directly interpretable time error
• 🗺️ Processed real logistics route & order pipelines
• 📦 Improved last-mile delivery planning accuracy

C++ Dijkstra's Algorithm Graph Theory DSA Priority Queue

Graph-based pathfinding engine for urban road networks

Structure    Weighted Adjacency List Graph
Algorithm    Dijkstra + Min-Heap Priority Queue
Complexity   O((V + E) log V)

• 🗺️ Weighted adjacency list for real city road networks
• ⚡ Dijkstra's Algorithm with priority queue for optimal speed
• 🔗 Realistic directional edge weight modelling
• 🚦 Globally optimal shortest paths between any two nodes
• 📈 Framework for smart city infrastructure planning

Python ML NLP Signal Processing Audio Features

Audio-intelligence engine using content-based filtering

Features    Tempo · Mood · Instrumentation · Key
Method      Content-Based Filtering + User Prefs
Stack       Python · Librosa · Scikit-Learn

• 🔊 Extracts tempo, mood, instrumentation & key from audio
• 🧬 Builds rich multi-dimensional music content profiles
• ⚖️ Hybrid: audio similarity + user preference weighting
• 📈 Advanced signal processing for robust feature extraction
• 🎧 Scalable recommendation engine architecture