01 / PROOF OF WORKTHREE SYSTEMS. THREE DIFFERENT CHALLENGES.
Selected engineering.
Practical, independently built projects spanning model development, retrieval-augmented generation, and the model lifecycle. Real methods. Measured results. Documented trade-offs.
01 / NLP & CLASSIFICATIONSupportSense
Banking intent intelligence
Pythonscikit-learnPyTorchTensorFlowGradio
SUPPORTSENSE MODEL / SVM
“My card hasn’t arrived yet.”
PREDICTED INTENTcard_arrival
77 INTENTS ↗
A customer-support classifier built end to end on BANKING77. Compared classical and neural approaches, then chose the simpler deployable model on test performance and operational trade-offs.
89.32%TEST ACCURACY
TF-IDF + LINEAR SVM
02 / GENERATIVE AI & RAGRAGAssist
Evidence-grounded document intelligence
LangChainFAISSFastAPIEmbeddingsGroq
DOCUMENT / 01
RESEARCH BRIEFAnswers backed
by evidence.↗ CITATION MATCH
An evaluated RAG system that retrieves passages, answers document questions and makes source evidence inspectable. Includes deliberate abstention logic and an honest error analysis.
100%RECALL@3
9-QUESTION EVALUATION SET
03 / MLOPS & SYSTEMSModelOps
Monitor. Evaluate. Retrain. Govern.
MLflowFastAPIStreamlitpytestGitHub Actions
MODEL HEALTH MONITORING
DRIFT CHECKPSI / KS
CHALLENGERREVIEW
RELEASEGUARDED ✓
An operational machine-learning platform covering drift checks, experiment tracking, controlled retraining, model serving, and an auditable champion–challenger release process.
GATED ↗CHAMPION–CHALLENGER
MODEL PROMOTION
A NOTE ON THE LIVE DEMOSPortfolio: instant. Project servers: prepared.
This website loads as static content. On a visit, it sends one low-priority request to each configured project backend to encourage free services to wake before you open them. Startup still depends on the hosting providers and is not guaranteed. No continuous keep-alive is used.
↗ ↗ ↗SEE THE ENGINEERING BEHIND THE WORK.
EXPLORE CAPABILITIES ↗