Data Scientist and AI Engineer. Currently building a no-code personal data analyst startup.

substack

projects

github

medium

gumroad

Cornellius Yudha Wijaya • Peerlist

Transform your data into actionable insights with ARIF, the AI-powered data analyst designed for marketers, researchers, and business professionals. No coding required, ARIF offers a comprehensive suite of features to help you understand and act on your data with confidence. <ul><li>Built-in analysis for marketing and research teams</li><li>Guided prompts for your objectives</li><li>No code analysis backed by statistics</li><li>Chat with your data for conversational insights</li><li>User-friendly interface that is 70% faster and 60% cheaper</li></ul> ARIF's platform provides AI-guided insights and storytelling, not just raw numbers. It supports a variety of data formats including Excel, CSV, PDF, and media files, allowing you to upload your data and gain insights in just three steps. <ol><li>Upload Your Data: Supports multiple formats</li><li>AI + Statistics: Enterprise-grade analysis in seconds</li><li>Gain Insights: Actionable intelligence for faster decisions</li></ol> With features like Sentiment Analysis, Customer Segmentation, and Trend Analysis, ARIF helps you uncover structured insights quickly and effectively. The platform ensures data privacy with robust encryption and compliance with PDPA, APP, and GDPR standards. Experience ARIF's capabilities with a free trial, offering 5 analyses or 1 month of access, whichever is earlier. No credit card required, and you can cancel anytime.

ARIF

Automated Machine Learning Explainability Pipeline is a production-ready framework that makes model decisions transparent, auditable, and compliant. It couples model-agnostic explainers with automation and observability so every run produces consistent, business-friendly evidence for why predictions were made. The system supports both global and local explainability: SHAP (Tree/Kernel) and LIME quantify overall feature influence and generate per-instance rationales, enabling bias checks, validation, and stakeholder trust. A typed data model (ExplainabilityReport) standardizes outputs—timestamps, model version, global importance, local attributions—and serializes them to JSON. Visual artifacts (SHAP summary bar plots) and LIME HTML explanations are created alongside an LLM-written executive summary (Gemini) that translates technical signals into plain language. All artifacts are logged to MLflow for lineage, comparison across versions, and long-term monitoring.

Automated Machine Learning Explainability Pip

The project is to analyze the Amazon Review Dataset to analyze which word (or word group) associated with the Positive, Neutral, or Negative sentiment. I also try to create a predictive model via One-VS-rest Logistic Regression as my main prediction model. 

Amazon Review Sentiment Analysis

RAG-To-Know is a project series that explores various RAG techniques, including implementation guides, use cases, and best practices. Each code is accompanied by an article designed to help researchers, developers, and enthusiasts understand and implement RAG systems efficiently.

RAG-To-Know

This project delivers a fully end-to-end ML system for email spam detection, taking a model from exploration to production with reliable monitoring and retraining. It starts with data profiling and lightweight feature engineering (text length, word counts), then benchmarks multiple classifiers and vectorizers (BoW, TF-IDF across Naive Bayes, Logistic Regression, Random Forest, XGBoost) using cross-validation, with MLflow tracking runs, metrics, artifacts, and models. Precision is prioritized to minimize false positives, and the selected pipeline (CountVectorizer + Naive Bayes) is serialized for serving. The production setup includes a FastAPI backend for inference and a simple Streamlit frontend for user input, both containerized. Docker Compose orchestrates services and mounts shared storage for models and data. Operational robustness is added via data drift detection with Evidently AI, comparing current inputs to reference distributions (labels, predicted probabilities, text descriptors). When drift is detected, an Airflow DAG conditionally triggers automated retraining inside a Dockerized environment, writes the updated pipeline back to the serving path, and keeps lineage clear. The result is a practical blueprint for small teams: reproducible experiments, metrics-driven model selection, one-command local deployment, and a scheduled governance loop that catches drift and refreshes the model without manual check which is ready to be lifted to cloud infrastructure when needed.

End-to-End Machine Learning Project Developme

The Resume Builder AI Agent is an intelligent resume-generation system powered by the ReAct agent framework. It uses large language models and job market insights to craft ATS-friendly resumes based on user-submitted profiles and (optionally) target job descriptions.

Resume Builder Agent

This project delivers a Multimodal Retrieval-Augmented Generation (RAG) system that answers complex questions by grounding responses in visual, audio, and textual evidence. It converts PDFs into images for vision processing, encodes images with CLIP, transcribes audio with Whisper, and embeds text using SentenceTransformer; all embeddings are indexed in ChromaDB for fast approximate nearest-neighbor search. At query time, the pipeline performs multimodal retrieval—fetching relevant images and transcript segments—then composes a unified context bundle that Qwen-VL can “see” and reason over. The model generates an explanation that references the retrieved artifacts while the UI surfaces both the answer and the supporting images/snippets for auditability. Designed for real workloads such as document Q&A, compliance reviews, meeting/audio search, and visual inspections, the architecture favors modularity (you can swap models or the vector store), GPU acceleration where available, and pragmatic engineering touches like chunking, batching, and caching to keep latency low. Optional re-ranking improves cross-modal relevance; simple guardrails filter unsafe inputs. The result is a practical, extensible blueprint for multimodal assistants: transparent enough for stakeholders, flexible enough for rapid iteration, and robust enough to move from prototype to production without rebuilding the core pipeline.