AI/ML Engineer building production-ready ML and Generative AI systems across modeling, serving, evaluation, monitoring, and decision support.
From validated data and model evaluation to deployed APIs and tools that help teams make better decisions.
| Area | What it means in practice |
|---|---|
| Production ML systems | Leakage-safe pipelines, calibrated predictions, threshold policies, reproducible artifacts, and outputs teams can act on |
| Generative AI & RAG | Retrieval evaluation, source attribution, hallucination-risk review, tool-calling agents, structured outputs, and quality checks before release |
| Evaluation & reliability | Evaluation for RAG, LLM, and agent systems, benchmark design, regression tests, trace review, and human review where needed |
| APIs & serving | Dockerized FastAPI services, strict schemas, versioned artifacts, and CI-backed delivery |
| Monitoring & ops | Telemetry, drift signals, cost and latency trade-offs, triage thresholds, and workflows that make review and handoff easier |
| Applied NLP & computer vision | Text classification, semantic search, threshold tuning, explainability, image restoration, and practical computer vision apps |
| Project | What it demonstrates |
|---|---|
| Fraud Risk Ops Platform | FastAPI inference and Streamlit analyst review with calibrated Random Forest and XGBoost models, validation F1 scores of 0.882 and 0.876, threshold policies, audit logs, batch jobs, and Prometheus/Grafana monitoring |
| RAG QA Command Center | CI-tested RAG evaluation across 3,824 QA runs and 93,375 retrieval events, covering retrieval diagnostics, hallucination-risk review, trace inspection, configuration comparison, and policy simulation |
| LLMOps Telemetry Command Center | Reliability, latency, cost, routing, and drift review across 9,000+ offline interactions, 1,595 sessions, and 438 user profiles, with triage thresholds and evidence exports |
| Advanced ML Sentiment Lab | TF-IDF pipelines, ROC-AUC/PR-AUC evaluation, threshold tuning, error review, live prediction, and exportable artifacts |
| Health Intelligence Platform | Behavioral risk analytics, cohort KPIs, threshold diagnostics, feature importance, trends, and scenario simulation |
| Old Photo Restorer | Gradio computer vision app with GFPGAN restoration, optional upscaling, before/after preview, and batch ZIP export |
| Project | Focus |
|---|---|
| Short-Video Intelligence Dashboard | Virality scoring, engagement metrics, creator leaderboards, timing patterns, and segment benchmarks |
| EV Charging Analytics | Geospatial infrastructure analytics, fast-DC allocation scenarios, market slices, and network planning |
| Football Matches Dashboard | European football and UCL analytics: KPIs, standings, team explorer, head-to-head, and interactive match tables |
| Seaborn & Matplotlib Visual Lab | Interactive Streamlit lab to build, compare, and export Seaborn vs Matplotlib charts with UI controls and generated code snippets |
| Hugging Face QuickStart Tool | Gradio tool that converts model/repo URLs into run commands, download snippets, file views, risk hints, and ZIP scaffolds |
| Project | Focus |
|---|---|
| Road Accident Risk Prediction | Two-stage risk scoring with LightGBM, XGBoost residual modeling, NNLS blending, stable OOF evaluation, and interpretable risk features |
| Cancer Risk Analysis | Synthetic tabular data, validation, leakage-aware benchmarking, and interpretable risk modeling for educational and analytical use |
| Clinical Deterioration Early Warning | 12-hour deterioration baseline with tabular models, probability ensembling, and cost-based threshold policy tables |
| Pima Diabetes Pipeline | End-to-end diabetes risk pipeline with EDA, feature engineering, calibration, cost-aware thresholding, and deployable artifacts |
| SMS Spam Detection | Dual TF-IDF pipeline with calibrated Linear SVM, nested CV, threshold tuning, explainability, and robustness checks |
| Text Sentiment Analysis | IMDB sentiment pipeline with calibrated TF-IDF baselines, threshold tuning, explainability, and BiLSTM baseline |
| Dataset | What it enables |
|---|---|
| RAG QA Logs & Corpus | RAG evaluation with QA logs, retrieval events, corpus documents, and evidence-style review workflows |
| LLM Production Telemetry | Offline LLMOps telemetry for reliability, latency, cost, routing, drift, and triage-policy review |
| Cancer Risk Factors | Synthetic health, lifestyle, environmental, and genetic features for leakage-aware risk modeling and educational analysis |
| Global EV Infrastructure | Standardized EV charging data for geospatial analytics, planning, and network modeling |
| YouTube Shorts & TikTok Trends 2025 | Short-form content analytics, trend exploration, creator benchmarks, and virality analysis |
| Digital Lifestyle & Mental Wellness | Behavioral signals for wellbeing analytics, cohort exploration, and predictive workflows |
| Category | Tools |
|---|---|
| Languages & Core |       |
| Data & Analytics |      |
| ML / DL |      |
| GenAI / RAG / Agents / CV |            |
| Apps & Interfaces |        |
| APIs & Serving |           |
| Evaluation, Monitoring & Quality |           |
- 🚀 Production ML & GenAI systems: FastAPI services, Dockerized delivery, evaluation workflows, and maintainable systems with clear handoff
- 🧠 RAG reliability: retrieval evaluation, grounded answers, source attribution, guardrails, regression tests, and trace review
- 🗂️ Validated data products: clean schemas, documented pipelines, reusable notebooks, and ML-ready artifacts
- 📊 Decision-support tooling: monitoring, threshold policies, operator-facing interfaces, and practical review workflows
Best contact: LinkedIn
If the work is useful, a ⭐ helps others find it.