These selected projects show the judgment that matters for business AI work: problem framing, source data, model choice, metrics, recommendation, deployment path, and human review. They are examples of how I think through AI, not a claim that every business needs a model.
Lead artifact
SuperKart does not stop at a notebook. That is the path a business can follow: data, model, interface, recommendation.
SuperKart compares regression approaches, selects a tuned Random Forest, serializes the pipeline, exposes a Flask API, wraps it with Streamlit, and ships through Docker-based Hugging Face Spaces.
R2 0.932 on the test set. MAPE 3.8%. Framed as a business view, not a leaderboard score.
Selective archive
Forecasting, targeting, decision support, maintenance risk, retrieval, and computer vision. The point is fit, not novelty.
Compared regression approaches, selected a tuned Random Forest (R2 0.932, MAPE 3.8%), serialized the pipeline, exposed a Flask API, wrapped it in Streamlit, and shipped through Docker-based Hugging Face Spaces.
model to API to UI 01
Unplanned generator failure is the expensive miss, so accuracy alone was not the right metric. Neural-network comparison with recall-first evaluation reached 0.837 test recall on the failure class.
recall-first evaluation 02
A visa certification model framed as prioritization support, not automated decision-making. Borderline and policy-sensitive cases stay in human hands.
responsible framing 03
Personal-loan targeting with customer segment review and outreach economics. ROC-AUC near 0.95, with income, education, CD accounts, and card spend as the top signals.
segmentation signal 04
A healthcare question-answering project grounded in medical manuals. Chunked source material, embedded with sentence transformers, retrieved top context, and compared answers against retrieved passages.
source-grounded answers 05
Helmet-detection exploration using a scratch CNN and VGG16 transfer-learning path. Treated as an exploration: the recommendation stays honest about what the metric table has and has not proven.
visual model exploration 06
The useful pattern
This is the bridge into consulting: AI only helps when it has source material, evaluation, deployment thinking, and human ownership.
Forecast revenue, prioritize outreach, flag risk, answer from manuals, or detect safety state. The business question comes first.
Regression, classification, neural network, computer vision, and RAG solve different problems. The shape follows the question.
RMSE, MAPE, F1, recall, ROC-AUC, and grounding scores tell different stories. The tradeoff is named, not hidden.
The best project connects model output to an API, UI, or clear business recommendation. A notebook that ships nothing has limited use.
Decision support, monitoring, retraining, and clear ownership. AI earns its place; it does not replace judgment.
What this signals
On framingThe business question comes before the algorithm. A well-chosen metric beats a better model on the wrong problem.
On deploymentA notebook that ships nothing has limited business value. The goal is always a path from model output to a decision someone can act on.
On AI in engagementsNot every business needs a model. AI earns its place when it reduces a real cost, speeds up a real decision, or flags a real risk.
A 20-minute Diagnostic Call. I will help name your top constraint and tell you honestly whether I can help. I take a few projects at a time, so each one gets real attention.
Start with a Diagnostic Call