█████╗ ██╗ ██╗ █████╗ ███████╗██╗ ██╗ ███╗ ███╗███████╗ ██╔══██╗██║ ██╔╝██╔══██╗██╔════╝██║ ██║ ████╗ ████║██╔════╝ ███████║█████╔╝ ███████║███████╗███████║ ██╔████╔██║███████╗ ██╔══██║██╔═██╗ ██╔══██║╚════██║██╔══██║ ██║╚██╔╝██║╚════██║ ██║ ██║██║ ██╗██║ ██║███████║██║ ██║ ██║ ╚═╝ ██║███████║Portfolio OS v2.0 — Type 'help' for commands
I build ML-powered systems that solve real problems — from fraud detection to explainable HR analytics. Certified by Oracle, Google, Meta & AWS. Optum STEM Scholar 2025–26. 4 live deployed apps.
I'm a Computer Science undergraduate specialising in Data Science, focused on turning raw data into actionable insight and practical solutions. I work with Python, SQL, Pandas, NumPy, Power BI, and ML frameworks.
My experience spans AI-based speech systems, data preprocessing pipelines, analytics workflows, and ML model deployment. I focus on understanding data deeply and solving real business problems — not just collecting certifications.
Actively preparing for Data Analyst, Data Science, and ML roles. Open to internships and entry-level opportunities. Also learning German 🇩🇪 on the side.
📧 ms29akash@gmail.com
Cost-sensitive ML pipeline weighting false negatives far higher than false positives. Real-time dashboard for live transaction scoring with SHAP explainability.
TF-IDF + semantic similarity engine accepting any dataset. Handles cold-start users and produces human-readable explanations per recommendation.
ML classification with SHAP-based explainability identifying at-risk employees. Interactive Streamlit dashboard bridging ML output and business decisions.
NLP + ML pipeline classifying job postings as legitimate or fraudulent. Paste any JD — get instant fraud risk score with SHAP feature attribution.
Flask-based web app using SpeechRecognition API supporting multiple Indian languages with real-time audio processing. Direct production deliverable.
Explainable AI-powered crop advisory system using NLP + ML that converts natural language into accurate, transparent crop recommendations with SHAP attribution per suggestion.
AI-powered decision intelligence system combining ML + LLM to forecast trends, detect anomalies, and generate actionable business insights. XGBoost + Random Forest + LLM reasoning layer.
Rigorous A/B testing analysis using statistical hypothesis testing, confidence intervals, and p-value evaluation to drive data-backed business decisions on conversion uplift.
Rule-based NLP chatbot serving as a smart farming companion. Provides crop guidance, weather advisory, and farming tips through an intuitive conversational interface.
A model with 99% accuracy that flags no fraud is useless in production. Standard ML treats a missed fraud (false negative) the same as a false alarm (false positive) — but in banking, a missed ₹50,000 fraud costs far more than wrongly flagging a transaction. The real objective isn't accuracy. It's minimizing cost-weighted error.
Extreme class imbalance required deliberate handling — not just oversampling. Applied SMOTE for minority class augmentation combined with class_weight='balanced' in the model. Features were PCA-transformed (V1–V28) plus Amount and Time. Amount required log-scaling; Time required cyclical encoding to capture daily patterns.
Prediction and explanation computed in a single forward pass. SHAP values cached per request. FastAPI handles async inference; Streamlit polls the API every 2s for live demo.
Used TreeExplainer (SHAP) — exact Shapley values for tree models, 10× faster than KernelExplainer. Each prediction outputs: which features pushed toward fraud, by how much, and the baseline fraud rate. This is what regulators and compliance teams actually need.