AI Deploy Network
AI AgentCross-Industry / Horizontal 11/16/2026

ML_Intern: Conversational AutoML Agent for Training Machine Learning Models

ML_Intern is a conversational AutoML agent that lets users upload a CSV, describe a prediction task, and automatically prepare data, train machine learning models, and evaluate results using more than 60 tools.

Hours Automated
0
Cost Savings
Currency not specified
Revenue Impact
Currency not specified

Business Challenge

Building a machine learning model usually involves several steps before training even begins. Someone has to inspect the data, handle missing values, prepare features, select an algorithm, tune parameters, and evaluate the result. These tasks can be difficult for people who understand their data but don't know how to write machine learning code. I wanted to make the process accessible through conversation. A user should be able to upload a spreadsheet, describe what they want to predict, and let the application handle the technical work. The main challenge was coordinating many machine learning operations while keeping track of datasets, trained models, and previous actions throughout a conversation.

Solution Delivered

I built ML_Intern, a conversational AutoML agent using Python, LangChain, OpenAI, Scikit-learn, FastAPI, and SQLite. Users upload a CSV file and explain their machine learning goal in plain English, such as predicting customer churn. The agent has more than 60 modular tools covering data inspection, exploratory analysis, cleaning, feature engineering, model selection, training, hyperparameter tuning, evaluation, and visualization. LangChain handles agent orchestration and tool selection. I implemented session state management so the system can keep track of datasets, transformations, models, and earlier decisions as users continue working. I also built a SQLite logging layer that records user requests and tool calls. This makes it possible to inspect what the agent did during a workflow and investigate errors. The application has a custom HTML, CSS, and JavaScript frontend connected to a FastAPI backend. I deployed ML_Intern on Hugging Face Spaces, where users can interact with the application through their browsers.

Outcomes Achieved

ML_Intern allows users to run machine learning workflows through natural-language instructions. A user can upload a CSV, describe a prediction problem, and have the agent perform data preparation, model training, tuning, and evaluation without manually writing the individual machine learning steps. The application combines more than 60 specialized tools into one conversational workflow. Session management preserves the context required to carry out several operations in sequence. The system can produce trained models, evaluation results, and visualizations. SQLite logging also provides a record of agent activity that can be inspected when debugging a run. I deployed the application on Hugging Face Spaces so the workflow can be demonstrated through a web interface. The project shows the technical feasibility of conversational AutoML. I have not yet measured time savings or model quality improvements against a manual baseline.

Measurable Business Outcome

ML_Intern integrates more than 60 machine learning tools into a single conversational agent. The system supports data inspection, cleaning, feature engineering, model selection, training, hyperparameter tuning, evaluation, and visualization. It maintains session state across multiple tool calls and records user requests and tool execution in SQLite for debugging. The application is deployed on Hugging Face Spaces and can produce trained machine learning models and evaluation reports from CSV datasets. The number of tools and supported workflow stages are implementation metrics. Time savings and comparative model performance have not yet been measured through a controlled benchmark.

Business Outcome Categories

Productivity ImprovementTime Savings