Summer Internship - II

Artificial Intelligence becomes more useful when it is connected with applications, data, and automated processes. During my summer internship, I explored this concept through AI automation using n8n.

The internship gave me practical experience with workflow automation, APIs, JSON, JavaScript, prompt engineering, cloud-based AI, and locally hosted Large Language Models. After learning the fundamentals of n8n, I developed two AI-powered automation workflows to apply these concepts to practical problems.

AI-Based Invoice and Receipt Processor

My first project focused on automating receipt processing. Manually reading receipts and recording details such as the vendor, date, amount, currency, and expense category can become repetitive and time-consuming.

I developed an n8n workflow that automatically detects receipt images uploaded to Google Drive and analyzes them using Google Gemini. The AI model extracts the required information and returns it in a structured JSON format.

I used JavaScript to clean and process the AI response before storing the extracted data in Google Sheets. After successful processing, the original receipt is automatically moved to a separate folder.

Prompt engineering was an important part of this project. Specific instructions were added to identify the final amount paid instead of subtotals, taxes, or individual item prices and to generate more consistent structured output.

AI-Based Research Paper Summarizer

My second project focused on automatically summarizing research papers using a locally hosted Large Language Model.

The workflow accepts a research paper in PDF format and extracts its textual content. The extracted text is sent to Llama 3.2, running locally through Ollama, using an HTTP API request.

The model generates a structured summary containing the Abstract Summary, Methodology, Results, Limitations, and Future Scope. Since the Ollama API returned multiple JSON response lines, I used JavaScript to parse and combine the generated text into one complete summary.

The final output is converted into a Markdown file and automatically stored in Google Drive. This project helped me understand how locally hosted AI models can communicate with workflow automation platforms through APIs.

Challenges and Learning

Developing these workflows also involved testing and debugging issues related to API limitations, JSON parsing, AI response formatting, local API communication, and file handling.

I learned that building an AI automation system involves more than simply connecting an AI model. The input must be prepared correctly, prompts must be designed carefully, responses must be processed, and the final output must be delivered to the appropriate application.

This internship strengthened my understanding of n8n, Google Gemini, Llama 3.2, Ollama, APIs, JSON, JavaScript, and prompt engineering. It also improved my debugging and problem-solving skills while giving me practical experience in building complete AI-powered workflows.

Final Thoughts

My summer internship was a valuable opportunity to explore the combination of artificial intelligence and workflow automation. The experience showed me how AI models can be integrated with applications and automated processes to reduce repetitive work and solve practical problems.

Working on these projects has increased my interest in AI automation, and I look forward to exploring more advanced workflows and AI integrations in the future.

— Aarya Shah