class Mohnish:
def __init__(self):
self.role = "Machine Learning Engineer"
self.education = "B.Tech ECE @ MIT ADT University, Pune (2023 - 2027)"
self.portfolio = "https://mohnish4code.github.io"
self.learning = ["MLOps", "FastAPI", "Docker", "GitHub Actions", "MLflow"]
self.interests = ["Computer Vision", "Anomaly Detection", "Time-Series"]
self.email = "mohnishshandilya.3000@gmail.com"
def motto(self):
return "Data -> Model -> Honest evaluation -> Deployed API"🚗 Vehicle Number Plate Detection System (click to expand / collapse)
Detects Indian number plates with a custom-trained YOLOv8n model, reads them with PaddleOCR (PP-OCRv4) and a plate-aware decoder that only accepts valid Indian formats (including BH-series and two-row plates), and logs confirmed plates to Google Sheets. Served by FastAPI with a mobile-friendly React app.
| 🎯 Detector mAP@50 | ✅ Exact reads (held-out) | 🔤 Character error rate | 🚫 Wrong reads auto-accepted |
|---|---|---|---|
| 0.994 | 85.3% (up from 54.4%) | 6.3% | 0 / 68 |
📈 Multi-Module Stock Prediction System (click to expand)
AlphaStream combines a forecasting model, technical rules and live FinBERT news sentiment in a decision engine that outputs one BUY / HOLD / SELL call for each of 10 NSE large caps.
- 📊 Compared ARIMA, XGBoost, LSTM and Transformer models (with and without news); XGBoost + FinBERT won and is served
- 🔒 Leakage-free walk-forward backtest: 6 one-year folds, 1,512 trading days per stock, 5-row embargo, against 4 baselines
- 🧾 Honest result: no model had a reliable next-day edge, and buy-and-hold beat them all. The README explains why
- 🧪 FastAPI + React + TypeScript, MLflow, Optuna, Docker, GitHub Actions and 79 pytest tests
🛡️ System Log Anomaly Detection (click to expand)
AIOps pipeline that turns 15.9M raw log lines (HDFS and BGL) into event templates with Drain, then flags abnormal sessions and time windows. Every alert links back to the log lines behind it.
| 🧩 Parsing accuracy (HDFS) | 🤖 Models benchmarked | ⚡ Streaming F1 (HDFS) | 🧪 Tests |
|---|---|---|---|
| 0.998 | 7 (rule → PCA → LSTM / Transformer) | 0.982 | 45 |
- 🔒 Chronological 60/20/20 splits; the deep models did not beat simpler methods on the test period, and the README explains why
📂 More projects
| Project | What it does | Stack |
|---|---|---|
| 💸 Expense Tracker | Budgets plus ML insights: unusual-expense alerts, overspend prediction, budget recommendations | FastAPI, SQLAlchemy, Scikit-learn, Docker |
| 🎯 ATS Resume Scanner | Scores a resume against a job description with an explainable 0–100 breakdown | FastAPI, Scikit-learn, TF-IDF |
| 📄 Question Paper Generator | Generates exam papers from course material with RAG | FAISS, Sentence Transformers, Google Gemini, Streamlit |
| 💳 Credit Risk Prediction | Predicts loan default risk | Scikit-learn, XGBoost, Streamlit |
| 📉 Customer Churn Prediction | Predicts which bank customers will leave | TensorFlow / Keras |
| 🎓 Student Performance Prediction | Predicts maths scores with a modular training pipeline | Scikit-learn, CatBoost, XGBoost, Flask |
🔍 See the full skill breakdown
| Area | Skills |
|---|---|
| 🧠 Machine Learning | Scikit-learn, XGBoost, Isolation Forest, Feature Engineering, Cross-Validation, Time-Series Forecasting, Anomaly Detection |
| 🔥 Deep Learning | TensorFlow / Keras, PyTorch, LSTM, Autoencoders, Transformers, YOLOv8, OpenCV |
| ✨ Generative AI | LLM APIs, RAG, Sentence Transformers, FAISS, Prompt Engineering |
| ⚙️ MLOps & Backend | FastAPI, REST APIs, Docker, GitHub Actions, MLflow, Pytest, Git |
| 📊 Data | Pandas, NumPy, SQL, EDA, Data Pipelines, Streamlit |
| 💻 Core CS | DSA, OOP, System Design, DBMS, Operating Systems, Computer Networks |
- 🎓 Kaggle: Intro to Machine Learning, Intermediate Machine Learning, Feature Engineering
- 🤖 Generative AI Value Added Course, iBase Technologies
⭐ If you like my work, drop a star on a repo, and feel free to reach out!