Noah Asano

AI-Powered Project Knowledge Manager (Gemini Observatory)

This project is a full-stack, AI-powered project knowledge manager developed during my Software Engineering Internship at NOIRLab’s Gemini Observatory. The goal was to reduce friction in how technical staff retrieve information across fragmented tools by providing a single, intelligent interface backed by LLM agents and a custom Model Context Protocol (MCP) server.

The application enables users to query, summarize, and interact with project data spanning operational logs, documentation, and issue tracking systems using natural language.

Project Overview

Gemini Observatory engineers and researchers rely on multiple platforms—such as Elasticsearch logs, Google Workspace, and Atlassian Jira/Confluence—to manage projects and operations. Retrieving context across these systems was time-consuming and required deep platform-specific knowledge.

I designed and implemented a system that uses LLM agents and a custom MCP server to unify these data sources into a single conversational and dashboard-driven interface, enabling faster decision-making and improved collaboration.

Architecture and Key Components

At the core of the system is a custom Model Context Protocol server written in Python. The MCP server exposes structured tools that allow LLM agents to safely and consistently interact with Gemini Observatory data sources, including:

A FastAPI-based REST backend handles OAuth 2.0 authentication, request routing, and communication between the frontend, LLM agents, and MCP tools. Credentials and tokens are securely stored in a PostgreSQL database using SQLAlchemy ORM.

LLM agents are responsible for orchestrating tool usage, synthesizing responses, and automating project management tasks such as summarization, cross-system lookup, and status reporting.

The frontend is a responsive web application built with HTML, CSS, and JavaScript, featuring a dashboard view and a chat-based interface that allows users to seamlessly interact with project data using natural language.

Impact

This system significantly reduced the time required for Gemini Observatory technical staff to locate and contextualize project information. By abstracting away platform-specific complexity and exposing a unified, intelligent interface, the application improved internal workflows and lowered the cognitive overhead of managing large-scale scientific projects.

Presentation and Communication

As part of a professional development program, I authored a formal project abstract, created a Google Slides presentation, and presented this work at an internal symposium. I was later selected to present the project at the National Science Foundation’s “The Solar System in Context” conference, where I communicated both the technical design and real-world impact to technical and non-technical audiences.

Source: Internal NOIRLab / Gemini Observatory project (not publicly available)