Back to resume

Portfolio projects

Browse 6 live demos and 1 narrative page. Each project opens on a focused same-origin portfolio route with clear demo expectations before you click.

Production ML Monitoring App

Live monitoring demo

MLOps application for sensor anomaly detection that compares candidate models, tracks experiments, serves inference, logs predictions, produces drift reports, and defines retraining triggers.

Submit sensor readings and inspect anomaly probability with monitoring context.

Python
FastAPI
W&B
Evidently
Docker
Predict an anomaly

RAG Evaluation and LLM Reliability

Live RAG demo

LLM evaluation harness for retrieval-augmented generation with deterministic retrieval, structured citations, a 60-question evaluation set, refusal checks, and cost and latency tracking.

Ask a grounded question and compare retrieval configurations.

Python
FastAPI
React
RAG
Evaluation
Try the RAG demo

Experimentation and Causal Inference Toolkit

Project write-up

Decision-support toolkit for product experimentation and causal inference, covering power analysis, A/B testing, CUPED, Bayesian and sequential testing, diff-in-diff, IPW, uplift modeling, and synthetic control.

Review the project goals, technical choices, and implementation context.

Python
Statistics
Causal Inference
React
View project page

Cloud Data Platform Mini-Lakehouse

Live data status

Analytics engineering mini-lakehouse that ingests FRED and GDELT data, builds raw-to-mart dbt models, runs quality checks, orchestrates refreshes, and publishes a BI-facing report.

Check the generated mart/report status from the portfolio facade.

Python
dbt
DuckDB
Prefect
BI
Check pipeline status

Job Role Classifier

Live classifier

Production-style classifier suite for job-posting role prediction with TF-IDF model comparison, deterministic synthetic augmentation, one-standard-error model selection, FastAPI serving, and a TypeScript dashboard.

Paste a job description and inspect predicted role probabilities.

Python
scikit-learn
PyTorch
FastAPI
React
Predict a job role

Black-Box Knowledge Distillation

Live distillation demo

Black-box distillation lab where a compact PyTorch student learns from teacher-labeled job postings, serves predictions through FastAPI, and reports fidelity, latency, confusion-matrix, and model-card artifacts.

Compare a compact student model against teacher-labeled job postings.

Python
PyTorch
FastAPI
React
Docker
Try student prediction

KV Cache Transformer Inference

Benchmark artifact

Transformer inference internals lab comparing naive decoder-only autoregressive generation with KV-cache generation, including cache-equivalence checks, decode timing, memory-growth analysis, and a static benchmark report.

Inspect KV-cache equality, decode timing, speedup, and cache-growth samples.

Python
PyTorch
Transformers
TypeScript
Inspect benchmark