IAToolkit Mini-Project: Scope, Phases, and Timeline

1. Overview

The IAToolkit Mini-Project is a structured, three-month initiative designed to deploy an internal AI assistant powered by IAToolkit and connected to real company data, documents, and workflows. It provides a practical, low-risk way for organizations to explore AI with tangible results and measurable progress.

Instead of aiming for a massive rollout, the mini-project focuses on a single business unit or client, ensuring that the scope remains manageable while delivering a production-grade assistant and deep organizational learning.

2. Goals of the Mini-Project

• Deploy a functional, company-specific AI assistant based on IAToolkit.
• Enable natural-language access to SQL databases and structured business information.
• Integrate internal documents using retrieval-augmented generation (RAG).
• Implement custom Python tools that allow the assistant to take real actions.
• Create a clean multi-tenant foundation for future assistants.
• Train internal teams on prompt design, schema modeling, and domain context.

3. Timeline and Phases (≈ 3 Months)

A complete mini-project spans three months, divided into four structured phases.

Phase 1 – Setup, Exploration, and Company Creation (≈ 3 weeks)

This phase establishes the foundation of the project by installing the toolkit, exploring its structure, aligning the scope, and creating the initial Company module.

• Install and run IAToolkit locally.
• Explore the sample_company and understand the architecture: prompts, schemas, context, tools.
• Define project scope and target questions with stakeholders.
• Create the new Company directory (e.g. companies/acme_corp/).
• Configure company.yaml with LLM, embeddings, tools, and branding.
• Connect SQL databases and validate the first generated queries.
• Set up basic authentication and internal testing access.

Phase 2 – Modeling Knowledge and Data (≈ 3–4 weeks)

This phase focuses on translating business knowledge into structured schemas and documents.

• Write YAML schema descriptions of the key tables and relationships.
• Add business rules and domain explanations in context/ as Markdown.
• Test LLM-generated SQL and iterate for accuracy.
• Refine terminology and structure with domain experts.
• Ensure reliable access to the datasets used in the first use cases.

Phase 3 – Tools, Workflows, and Documents (≈ 3–4 weeks)

Once the assistant understands your data, the next step is enabling it to act and reason.

• Create reusable prompts for recurring business tasks.
• Implement Python tools (reports, notifications, calculations, API integrations).
• Define tool schemas in company.yaml for structured function calling.
• Load internal documents into the vector store for RAG.
• Validate multi-step workflows combining SQL, RAG, and tools.

Phase 4 – Pilot and Production Rollout (≈ 2–3 weeks)

The last phase validates the assistant with real users and prepares it for production.

• Deploy a pilot version to a controlled user group.
• Monitor usage using IAToolkit’s history and feedback tools.
• Refine prompts, schemas, and tool behavior based on real-world interactions.
• Finalize production deployment and internal documentation.

By the end of the three-month cycle, the organization has both a working AI assistant and the internal knowledge to continue expanding it to new datasets, workflows, and business units.

4. Expected Outcomes

• A fully functional AI assistant integrated with your real business data.
• A documented set of schema files, prompts, tools, and context documents.
• A maintainable Company module ready for expansion.
• Hands-on experience with prompt engineering, SQL modeling, and RAG.
• A clear roadmap for scaling AI to additional business units.

5. Team and Technical Requirements

Team

• 1 Python Developer (responsible for configuration and tools)
• 1 Domain Expert (validates answers, defines rules, provides examples)
• Optional: DevOps engineer for cloud setup and monitoring

Technical Stack

• Python 3.12+
• PostgreSQL for IAToolkit’s internal database
• Redis for session handling
• Local or S3-compatible object storage for documents
• A Git repository for your Company code and configuration
• A cloud environment capable of hosting a Flask application

6. Why This Approach Works

The three-month structure provides the right balance: long enough to deploy something meaningful, but short enough to stay focused and manageable. It ensures controlled experimentation, real-world validation, and internal learning before scaling to larger initiatives.

“The goal is not only to deliver an assistant, but to help the organization understand how to design and operate AI systems grounded in their own data.”

7. Next Steps

Because IAToolkit is fully open source, organizations can begin immediately:

• Clone the repository and experiment locally.
• Use this mini-project structure as an implementation guide.
• Adapt schemas, prompts, and tools to your own data and workflows.

If you're exploring AI for internal use, the IAToolkit Mini-Project offers a clear, practical, and structured way to get started.