Machine Learning Engineer | AI Architect | Quantitative Researcher
ML Engineer with 1+ year of production experience shipping end-to-end systems. Expertise includes edge-deployed inference engines on Raspberry Pi, LLM pipelines on serverless Cloudflare Workers, and quantitative research frameworks validated across 10,000 Monte Carlo simulations. This website serves as a live portfolio showcasing my technical work across ML, AI, and full-stack development.
| METR — Market Exposure Timing Research | [GitHub] [Live] |
|---|---|
| Python, Polars, XGBoost, scikit-learn |
- Constructed a two-layer meta-labeling framework on 3,440+ days of multi-asset Indian market data (Nifty 50, GoldBees, USD/INR).
- Validated against 10,000 Monte Carlo simulations. The GoldBees filter achieved a Sharpe 1.48 and Calmar 1.87 with a 66.1% OOS win rate.
- Applied Fractional Differentiation (d=0.45) preserving 81% memory with stationarity, Triple Barrier labeling, and SHAP feature importance analysis.
| Network IDS Detection Engine | [GitHub] [Live] |
|---|---|
| LightGBM, XGBoost, PyTorch, FastAPI, cuML |
- Engineered a dual-detection intrusion detection system (signature-based + unsupervised anomaly detection) on 2.8M rows of network traffic (CICIDS2017).
- Benchmarked LightGBM at 99% accuracy (F1: 0.99). Reduced feature space 51% (69 to 34 features) via IncrementalPCA while retaining 98.97% variance.
- Deployed as a FastAPI service on Raspberry Pi 5, exposed publicly via Cloudflare Tunnel.
| CaptchaOCR | [GitHub] [Live] |
|---|---|
| Python, PyTorch, CNN + BiLSTM, CTC Loss |
- Trained a recognition system achieving 96%+ character-level accuracy on distorted sequences via synthetic training corpus generation.
- Published to Hugging Face Spaces for public model deployment and inference.
- Languages: Python (Advanced), TypeScript, C++
- ML & Modeling: Supervised Learning, SVMs, LightGBM, XGBoost, CNN, BiLSTM, Time-Series Forecasting, Anomaly Detection, Feature Engineering, Cross-Validation
- LLMs & GenAI: LLM Integration (GPT-4, Gemini), Multimodal Pipelines, Model Context Protocol (MCP), RAG
- MLOps & Deployment: FastAPI, REST APIs, Docker, GitHub Actions, MLflow, Cloudflare Workers, Cloudflare Tunnel, Edge Deployment (Raspberry Pi 5)
- Frameworks: TensorFlow, PyTorch, scikit-learn, Hugging Face, Pandas, NumPy, Polars, PostgreSQL
HumanizeIQ | AI Intern
- Architected MCP servers on serverless Cloudflare Workers for multimodal document workflows.
- Designed recruiter call analysis pipelines integrating LLMs, eliminating 100% of manual post-call processing.
- Built REST APIs for background job orchestration and scalable asynchronous processing.
Jaypee Institute of Information Technology B.Tech in Electronics and Communication Engineering (2022 – 2026)
- Relevant Coursework: Machine Learning (A+), Deep Learning (A+), Advanced Statistics (A), Multivariate & Differential Calculus.