Professional Summary
AI engineer who ships clear, runnable examples that scale from local notebooks to production-ready baselines. Practical focus on LLM orchestration, anomaly detection (visual & tabular), causal inference, and feature pipelines. I build lean, reproducible AI systems with strong observability, reproducible experiments, and fast feedback loops — balancing model quality with operational simplicity.
Languages: Python, SQL
AI & ML Frameworks: LangGraph, LangChain, PyTorch, Lightning, scikit-learn, PyOD
Experimentation & Repro: MLflow, Jupyter, deterministic seeds, structured configs
Data Quality & Catalog: Great Expectations (GX), OpenMetadata
Serving & Observability: BentoML, Prometheus, Grafana, Streamlit
Data Engineering & Scale: PySpark, Spark SQL, Docker, MySQL, PostgreSQL
Optimization & Simulation: OR-Tools, python-constraint, SimPy, Salabim
- linkedin_article_using_lang_graph — Example of using LangGraph for multi-step LLM orchestration and stateful workflows. Automates generating, refining, and summarizing long-form LinkedIn content using a graph-based reasoning engine. Tech: LangGraph, LangChain, Python, LLMs.
- fairml-demo — Notebook-based exploration of Fairness in ML, showing bias detection and model parity checks with explainability overlays. Tech: scikit-learn, Fairlearn, SHAP.
- granger_casuality_test — Implementation of Granger causality tests with rolling window evaluation and visualization to identify directional dependencies between time series. Tech: statsmodels, pandas, matplotlib.
- survival_analysis_bert — Survival analysis using transformer embeddings (BERT) for feature extraction and Cox proportional hazards modeling. Tech: PyTorch, lifelines, transformers.*
- langflow-demo — Demo of building LLM apps with Langflow: chaining prompts, integrating APIs, and deploying lightweight workflows. Tech: Langflow, Python, LLMs.*
- hospital-queue-simulation — Discrete-event simulation of patient flow with scenario builder to explore wait-time vs staffing cost. Tech: SimPy/Salabim, Streamlit.*
- pyspark-optimization — Practical Spark performance patterns (partitioning, joins, AQE, memory) with a tuning checklist. Tech: PySpark, Spark SQL.*
- python-constraint-programming — Scheduling/assignment/routing recipes with reproducible OR-Tools and python-constraint notebooks. Tech: OR-Tools.*
- bentoml-graphana-demo — Model serving with BentoML plus Prometheus metrics and Grafana dashboards for latency/throughput. Tech: BentoML, Prometheus, Grafana.*
- local-llm-paper-summarizer — Local pipeline to fetch PDFs, summarize, extract highlights, and export concise notes. Tech: Python, LLM runtime (local).*
| Repository | Focus | Tech Highlights |
|---|---|---|
| openmetadata-docker-demo | Local OpenMetadata stack with ingestion, profiling, and quality tests | Docker, YAML, MySQL |
| federated_learning_flower | Minimal federated learning demos | Flower, Python |
| visual-anomaly-detection | Image anomaly detection with overlays/heatmaps | Jupyter, PyTorch |
| pyod-anomaly-detection | PyOD baselines across tabular/text/graph/time series | PyOD, scikit-learn |
| automl-benchmark | Compact AutoML baselines with repeatable splits | Jupyter, sklearn |
| timeseries-analysis | Rolling stats, features, detectors | Pandas, statsmodels |
| ML-Flow | Experiment tracking patterns & reproducible runs | MLflow, Python |
| llm-metadata-generator | Scripts to curate metadata for LLM tasks | Python |
| llm-lazy-recipe-generator | Small LLM utility; MIT-licensed | Python |
- Reproducible, download-friendly datasets and clear READMEs
- Side-by-side visuals (e.g., normal vs anomalous) to build intuition
- Simple baselines first, then measurable improvements
- Preference for modular, instrumented workflows with observability built-in
GitHub: github.com/monika393