How to Build a Custom Project in Claude | Algorithms Tutor Example | Humanitarians AI

Chatting with an AI assistant normally means starting from zero every time: no memory of what you told it last session, no fixed tone, and a real risk of it drifting off topic or making things up. Claude's Custom Projects feature is built to fix that, and this video walks through building one from scratch using a friendly, concrete example: a personalized algorithms tutor.

Why a custom project instead of a regular chat

Large language models are powerful, but they tend to forget things easily, lose user context, or lack domain memory, and they can sometimes state things confidently that aren't true. A custom project addresses that by letting you define upfront what the AI should know, remember, and how it should behave, rather than re-explaining context in every conversation.

Creating the project

The process starts from the project options on the Claude interface. Creating a new project just needs a title and description, in this case a "personalized algorithms tutor" meant to explain data structures and algorithms concepts in simple terms. After the project is created, its details can be edited any time through the project's settings.

Writing instructions that shape tone and role

The most important step is adding instructions, the text that actually tailors how Claude responds inside that project. For the algorithms tutor, the instructions assign Claude the role of a personalized algorithms tutor, set the objective as explaining concepts and solving problems step by step, and set the tone as friendly, encouraging, and student-focused. These instructions can be customized freely depending on what a given project needs to do.

Uploading files for grounding

Beyond instructions, a project can also take uploaded files, PDFs, code, notes, anything that gives Claude additional material to draw on. In this example, PDF and Python files containing example code were uploaded to ground the tutor's answers in real material rather than general knowledge alone. Notably, a substantial set of files only used about 11% of the available project capacity, meaning a lot of supporting material can be added before running into any limits.

Testing the result

With instructions and files in place, the tutor was tested with a simple question: explain the concept of a single linked list. The response came back matching the tone set in the instructions, friendly, step-by-step, and appropriately simple, confirming that the project setup was doing what it was designed to do.

Key takeaways

• A Claude Custom Project lets you define role, tone, and memory upfront instead of re-explaining context in every chat.
• Creating a project just requires a title and description to get started.
• Instructions are the key lever for shaping tone, role, and teaching style, and they can be edited any time.
• Uploaded files (PDFs, code, notes) ground the assistant's answers in your own material rather than general knowledge alone.
• A meaningful set of uploaded files can use a surprisingly small share of the available project capacity.
• Testing with a real question is the fastest way to confirm the instructions are actually shaping the assistant's behavior.

Who this is for

This is a practical, no-coding-required tutorial for educators, developers, or anyone building a specialized AI assistant, whether for teaching, documentation, or research, and it's a useful starting template from Humanitarians AI for anyone building their first Claude Custom Project.