Скачать или смотреть 5-Day AI Agents Intensive Course, Capstone Project Electoral Roll Anomaly Detector

How My Code Works (The Logic Behind the Agent)
Since I’m not a professional developer, I wanted to build something that was powerful but used standard, reliable tools. I moved away from experimental libraries and built this using the official Google Generative AI SDK in Google Colab.

Here is the simple logic flow of my code:

The Setup: First, I install the standard Google libraries (google-generativeai) and pandas for data handling. I securely load my three API keys (AI, Search, and CSE ID) from the Colab "Secrets" vault so they are never exposed in the code.

The "Knowledge Base": I create a voters.csv file right inside the code. This mimics a real government database. I purposefully hid anomalies in it—like a "fake family" with impossible ages, duplicate photos, and a suspicious bulk address.

The "Eyes" (Custom Tool): This is the most important part. I wrote a python function called Google Search_tool. This isn't just a script; it’s a tool I give to the AI. It allows the agent to pause, go to the internet, verify an address, and bring that information back to answer my question.

The Agent: I initialize the Gemini model (gemini-1.5-flash-latest). In the "System Instructions," I injected the CSV data and gave it strict rules: Answer from the data, but if you see a suspicious address, you MUST check it on Google.

The Audit: Finally, I run a loop of 4 specific prompts. The agent reads the data, finds the ghosts, and uses the search tool to prove that "1, Small Apt, Kochi" is likely too small for 3 unrelated voters.