Welcome to my centralized research hub. This repository contains specialized projects focusing on Multi-Modal Risk Fusion, Transformer Architectures, Explainable AI (XAI), and Strategic Market Analytics.
My work focuses on bridging foundational machine learning research with actionable corporate strategy, emphasizing high-fidelity diagnostics and evidence-based decision-making.
| Rank | Project Name | Research Category | Primary Tech Stack | Repository Link |
|---|---|---|---|---|
| 1 | Sentinel Gold | Multi-Modal Risk Fusion | Spark, Kafka, GPT-2 | View Repo |
| 2 | ELS-Pulse | Heuristic NLP & Brand Audit | Ensemble ML, Streamlit | View Repo |
| 3 | InsuraPulse | Explainable AI (XAI) | SHAP, Scikit-Learn | View Repo |
| 4 | QuantPro | Signal Processing | yfinance, Pandas | View Repo |
| 5 | MoviePulse | Latent Factor Modeling | SVD, NLP, TF-IDF | View Repo |
| 6 | RetailPulse | Association Theory | Apriori, FP-Growth | View Repo |
| 7 | Consumer Audit | Consensus Methodology | Flask, Scikit-Learn | View Repo |
| 8 | LLM Foundations | Transformer Theory | PyTorch, Tiktoken | View Repo |
| 9 | Airline Fleet Planning | Strategic CAPEX Analysis | Power BI, Power Query | View Repo |
| 10 | Shopping Trends | Behavioral Economics | Power BI, DAX | View Repo |
- Strategic Problem: Modern security suffers from disconnected digital threat feeds and physical surveillance silos.
- Research Solution: A fusion engine correlating high-velocity Kafka indicators with live CCTV telemetry via PySpark.
- Outcome: Leverages GPT-2 to generate automated executive summaries, bridging the gap between technical telemetry and management-level reporting.
- Strategic Problem: Linguistic nuances like sarcasm cause "sentiment flipping," leading to high error rates in automated brand audits.
- Research Solution: A Voting Ensemble (87.1% accuracy) utilizing custom sarcasm-aware heuristics to identify linguistic nuances.
- Outcome: Provides high-fidelity sentiment auditing through Neural Probability Heatmapping and TF-IDF feature significance analysis.
- Strategic Problem: The "Black Box" nature of complex regression models prevents actuarial transparency and stakeholder trust.
- Research Solution: Applied SHAP (SHapley Additive exPlanations) to isolate and interpret non-linear drivers in medical premiums.
- Outcome: Delivers transparent, auditable models by normalizing skewed residual patterns through Log-Linear transformations.
- Strategic Problem: Financial research requires low-latency pipelines capable of extracting signals from high-velocity data.
- Research Solution: A robust pipeline utilizing real-time yfinance data to monitor market volatility and price-action correlations.
- Strategic Problem: Verifying the mathematical mechanics, convergence stability, and latent organization of modern LLM architectures.
- Research Solution: A ground-up PyTorch implementation covering BPE tokenization, Multi-head Self-Attention, and Layer Normalization.
- Validation: Verified through mathematical monitoring of Perplexity and Loss, ensuring foundational architectural stability.
- Modeling: Transformer Architecture, Matrix Factorization (SVD), Ensemble Learning, BPE Tokenization.
- Statistics: Explainable AI (SHAP), Z-Score/Lift Analysis, Neural Probability Mapping, Feature Significance (TF-IDF).
- Engineering: Apache Spark, Kafka, Real-time APIs (yfinance), Power Query (M/ETL), Flask.
- LinkedIn: www.linkedin.com/in/mittal-chauhan
- Email: mittalchauhan2903@gmail.com
- Portfolio Note: Each sub-directory includes a detailed README and development notebooks (
.ipynb) documenting the full research methodology.