Skip to content

Kunal-Somani/n8n-agents-project

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AI Agent Development Portfolio

This repository contains a portfolio of specialized AI workflow automation agents built using n8n and robust Large Language Model (LLM) and Retrieval-Augmented Generation (RAG) architectures.

The core objective is to demonstrate proficiency in creating robust, autonomous systems that integrate external APIs with local/cloud-based generative models for complex decision-making.


Core Infrastructure and Design Philosophy

We leverage a hybrid architecture combining local open-source models with cloud-based intelligence to achieve stability and high-quality results.

1. Platform & Orchestration

  • n8n (Self-Hosted): Serves as the central nervous system (orchestration layer). We use Docker Compose for reliable, self-hosted deployment.
  • Demonstrated Skill: Managing complex, multi-step agent logic, handling data flow reliability, and creating robust control flows (e.g., using the Switch node).

2. Reasoning & Context (LLM & RAG)

  • LLMs Used: Google Gemini 2.5 Flash (for high-precision, instruction-following reasoning) and Ollama (Mistral, Gemma, Nomic) for local processing.
  • RAG: Implemented using Ollama's Nomic-Embed-Text and the built-in Simple Vector Store to ground AI responses in project-specific documentation, effectively eliminating hallucination.

3. Action & Integration

  • Agent Tool Use: Agents are designed to utilize Tool Use capability (calling other systems).
  • Custom API Integration: We use the HTTP Request node to call and interact with custom APIs (like GitHub's REST API), transforming passive reasoning into autonomous action.

Projects Created

Project 1: GitHub Issue Triage Agent

An autonomous agent that monitors a designated repository for new issues, automatically classifies them, and provides immediate, context-aware action.

  • Primary Goal: Automated issue classification and initial response, ensuring rapid and accurate triage without human intervention.
  • Key LLM Function: Employs the Gemini 2.5 Flash model for superior classification logic (question, bug, or feature), utilizing a dedicated multi-step workflow to ensure model reliability.
  • RAG Implementation: A dedicated Answer Agent is instantiated only on the question path, using a RAG pipeline to retrieve relevant installation or usage instructions from project documentation before drafting a final comment.
  • Actionable Integration: Executes real-world actions via the GitHub API for two distinct purposes: posting the RAG-generated answer and applying the corresponding triage label.

Project 2: Smart Scheduled Research Assistant

An efficient, autonomous pipeline designed to monitor the academic landscape and deliver curated intelligence directly to the user's inbox on a fixed schedule.

  • Primary Goal: Automated data acquisition, complex data transformation, LLM synthesis, and report delivery in a resource-efficient manner.
  • Key LLM Function: Employs Gemini 2.5 Flash for advanced synthesis of complex abstracts, distilling multiple papers into a single, high-level digest.
  • RAG Implementation: Uses a dynamic RAG pipeline that ingests the latest research abstracts upon every scheduled run, guaranteeing the summary is grounded in the current day's findings.
  • Actionable Integration: Demonstrates multi-stage custom API integration including fetching complex Atom XML data from the ArXiv API and utilizing SMTP to send the final report.

Project 3: Live Hacker News 'Sentiment' Agent

An advanced, scheduled agent that performs live data aggregation from the Hacker News API and analyzes unstructured, real-time user-generated content.

  • Primary Goal: To demonstrate a dynamic RAG pipeline on live user comments, moving beyond static document analysis to perform real-time sentiment synthesis.
  • Key LLM Function: Employs Gemini 2.5 Flash to synthesize sentiment and key discussion points from dozens of unstructured user comments into a concise, executive-level summary.
  • RAG Implementation: This is the project's core complexity. It creates a "just-in-time" vector store for each individual news story, using the story's title as a dynamic Memory Key. This perfectly isolates the AI's context for each summary.
  • Actionable Integration: Fetches data via a complex, nested-loop API call (story -> comments) and delivers a single, aggregated report via SMTP.

Project 4: Unstructured Text to Google Sheets Agent

A high-value business automation agent that acts as an automated data-entry specialist, demonstrating a "unstructured-to-structured" pipeline.

  • Primary Goal: To parse unstructured text (like an email), extract key entities, and write them to a Google Sheet, proving efficiency in data integration.
  • Key LLM Function: Employs Gemini 2.5 Flash for structured data extraction (text-to-JSON).
  • RAG Implementation: Uses a "RAG-Zero" or "In-Context Learning" approach by providing "few-shot" examples in the system prompt, which is a highly effective and low-overhead RAG pattern.
  • Actionable Integration: Uses the Google Sheets API (via Service Account) to authenticate and append the structured data as a new row in a spreadsheet.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors