A data science operating system for no-code and low-code experimentation, cleaning, and deployment.
DS-Forge turns common ML workflows into a full-stack platform with spreadsheet-style cleaning, rich feature engineering, model training, and auto-generated inference APIs.
The frontend-led identity of DS-Forge comes from a TypeScript and Next.js surface paired with a Python ML backend. That split helps the platform feel structured and productized instead of notebook-driven.
DS-Forge is built as a no-code and low-code studio that covers ingestion, cleaning, feature engineering, training, evaluation, and deployment.
The stack pairs Next.js with FastAPI, Pandas, NumPy, Scikit-Learn, and Docker, giving it a strong split between interface and data-engine responsibilities.
A key design idea is exact pipeline preservation so the same transformations used in training are reapplied during inference.
Supports CSV, JSON, XLSX, and text-based ingestion with schema-aware parsing.
Provides 25+ atomic cleaning operations plus manual data-grid editing when automation is not enough.
Surfaces smart recommendations so data cleaning becomes more guided and less trial-and-error.
Includes 28+ feature transformations spanning dimensionality reduction, encoding, and preprocessing.
Exposes a curated model zoo for regression and classification tasks.
Moves trained models directly into REST-style deployment with a live inference playground.
The manual grid, smart recommendations, and one-click API deployment are some of the most product-defining parts of the platform.
The system is also built to run cleanly as separate frontend and backend containers.
Implemented a manual data grid with winsorization, outlier removal, and smart recommendations.
Built a feature engineering suite covering PCA, t-SNE, Isomap, polynomial features, and robust scaling.
Auto-deployed trained models as REST APIs with inference-time encoding preservation.
Upload data, validate schema, and clean it through targeted or global operations.
Apply transformations, train or compare models, then evaluate and iterate.
Deploy the selected model as an API while preserving the training-time preprocessing chain.
Operational completeness is a major part of the platform, not only model training.
The containerized workflow also makes the system easier to reproduce, demonstrate, and deploy.
DS-Forge is centered on a simple product idea: the most common machine-learning tasks should feel like one continuous workflow instead of a handoff between notebooks, scripts, and serving code. Data cleaning, feature engineering, model training, evaluation, and deployment are presented as parts of the same operating environment. That makes the platform easier to understand for fast experimentation while still keeping the output operationally useful.
A major design choice was to make deployment a native part of the product rather than an afterthought. Once a model is trained and selected, the same workflow can move it into API-backed inference while preserving the preprocessing chain used during training. That continuity is important because it closes the gap between experimentation and actual use.