Available for enterprise deployments

Case studies & engineering projects.

Production-grade engineering solutions, infrastructure optimizations, pipeline designs, and robust AI implementations deployed to solve real-world complexity.

01Read case ↗
Data Engineering & BI

Hospital Data Warehouse & CDC-Driven BI Pipeline

Designed and built a centralized data warehouse and BI pipeline for a healthcare provider — consolidating fragmented operational systems into a single source of truth and keeping it continuously synchronized via Change Data Capture (CDC).

Data WarehouseETL PipelineCDCStar / Snowflake
Key impact
Unified scattered hospital systems into one analytics-ready warehouse; CDC keeps it in near-real-time sync. Pre-production, preparing for go-live.
02View on GitHub ↗
Agentic AI & Automation

Autonomous Finance Analyst Agent

An agentic system that autonomously composes and delivers daily financial briefings — fetching stock quotes and news for a watchlist, analyzing them, and sending to Telegram on a schedule with no human in the loop.

LangGraphFastAPIn8nGPT-4o-miniDocker
Key impact
End-to-end agentic pipeline (planner → agent ⇄ tools → finalize) with per-run cost tracking; read-only by design for safe unattended runs.
03View on GitHub ↗
Generative AI & Data Eng

Enterprise Multi-Format Document Intelligence & RAG Pipeline

A production-grade document processing and insight extraction engine featuring custom multi-format parsing, context-aware semantic chunking, vector search, and cross-encoder reranking.

FastAPIPythonVector SearchRerankerPostgreSQL
Key impact
Automated structured text/table extraction from complex assets (PDF, DOCX, XLSX, PPTX) via modular microservices, minimizing retrieval noise.
04Read case ↗
Enterprise AI & Infra

Cloud API to Local Inference Migration

Eliminated reliance on external third-party APIs by deploying a localized open-source model (Qwen) on private cloud infrastructure to securely process medical records.

QwenOllamavLLMHPE PCAIPython
Key impact
Slashed response latency from 3.6s to 1.1s (69% lift) while ensuring 100% compliance with strict data-privacy regulations.
05Read case ↗
Generative AI & Architecture

On-Premise RAG Architecture & Clinical Document Analysis

A secure Retrieval-Augmented Generation pipeline for automated EMR summarization and billing analysis without exposing sensitive patient data to external networks.

LangChainn8nPostgreSQLC# / .NET Core
Key impact
Enabled 100% secure, fully on-premise processing of highly confidential hospital documents.
06View on GitHub ↗
MLOps & Data Engineering

Multimodal MLOps Pipeline for Clinical Risk Assessment

A 10-container microservices topology using Apache Airflow chunked DAGs and MLflow to process multimodal clinical data and prevent out-of-memory failures.

Docker ComposeAirflowMLflowFastAPIXGBoost
Key impact
Elevated predictive performance from a 0.73 tabular baseline to a 0.9198 multimodal fusion AUC, achieving an elite 90.9% recall.
07View on GitHub ↗
Data Science & Pipelines

Production-Grade Price Intelligence & Anomaly Detection

A multi-stage pricing recovery system that accurately reconstructs missing e-commerce data during scraping outages using cyclical temporal encodings and historical value ratios.

LightGBMOptunaPythonFeature Engineering
Key impact
Reconstructed missing product prices during scraping outages with multilevel fallback defenses, maintaining 100% mathematical stability in production.
08Read case ↗
Biosignal & Deep Learning

Pulmonary Arterial Hypertension Detection from PPG Signals

Led an end-to-end signal processing and classification pipeline using Wavelet Scattering Transform and a Bidirectional LSTM for non-invasive PAH screening.

PythonTensorFlowBi-LSTMWavelet Transform
Key impact
Awarded 1st Place at the West Java Health Tech Innovation Competition 2025. Potential to push survival rates past 80-90%.
09View on GitHub ↗
Generative AI & Full-Stack

End-to-End RAG Chatbot for Academic Thesis QA

An interactive, full-stack chatbot that uses natural-language queries to extract insights and methodologies from a multi-page biomedical thesis.

LangChainChromaDBGroq APIStreamlitDocker
Key impact
Automated targeted knowledge extraction; fully containerized and publicly hosted on Hugging Face Spaces.
10Read case ↗
ML & Clinical Research

Type 2 Diabetes Prediction Using Vascular Stiffness Parameters

A comparative machine-learning study using a Leave-One-Out Cross-Validation strategy to identify T2DM from non-invasive vascular and hemodynamic parameters.

Random ForestXGBoostSVMScikit-LearnLOOCV
Key impact
Presented at ISPACS 2025. Random Forest achieved 85.5% overall accuracy and a 0.921 AUC.
11View on GitHub ↗
Computer Vision & Imaging

Brain Tumor MRI Classification via Optimized 3-Layer CNN

A deep-learning CNN architecture to classify over 7,000 high-resolution brain MRI scans across varied clinical points of view.

PythonTensorFlowKerasCNNMatplotlib
Key impact
Achieved 96.01% validation accuracy while maintaining high model generalization across distinct anatomical planes.
12View on GitHub ↗
Computer Vision & Automation

Industrial Color Detection & Quality Control System

An automated computer-vision pipeline for color segmentation and cap classification on an industrial manufacturing line.

PythonOpenCVPoetryDockerCLI
Key impact
Boosted accuracy from 75.2% to 92.7% by identifying and disabling destructive color-jittering augmentations.