Akash Pandey

Jan 2026 – Present

Privacy-safe synthetic data · full statistical fidelity

A CTGAN-based generative AI system that synthesizes privacy-safe training data while preserving the source's full statistical fidelity. Production-grade pipeline for ingestion, automated cleaning, imputation, and hyperparameter optimization, with an auditable validation layer — KS test, Total Variation Distance, and ML efficacy via train-synthetic-test-real. Shipped as a full-stack Streamlit app backed by Supabase (row-level security), with CI/CD via GitHub Actions and unit tests.

~99K orders · 7-table star schema · 4-page BI dashboard

End-to-end analytics on ~99,000 orders from the Olist Brazilian e-commerce dataset. Modeled 7 relational tables into a star schema (plus a custom Date dimension) in PostgreSQL, then answered core questions on sales, retention, delivery, and payments across a 4-page interactive Power BI dashboard. Surfaced insights like late deliveries dropping average review score from 4.3 → 2.57, and only ~3% of customers making a repeat purchase.

Cleaned & published on Kaggle

Exploratory data analysis on raw TMDB movie data to surface quality issues, inconsistencies, and missing patterns — then cleaned, deduplicated, and standardized the dataset and published it publicly on Kaggle for other practitioners to build on.

DeepLearning.AI · Andrew Ng (Coursera)

Supervised & unsupervised learning, advanced algorithms, and neural networks.

Oracle

Oracle Cloud Infrastructure AI certification — scored 85%.

NPTEL

Elite + Silver medal · scored 79%.

Google Cloud

Cloud AI / ML foundations.

Business Club, IIT Delhi

Ranked #67 among participating teams.

HackerRank

Certificate of Achievement.