ChippulseAI

Agentic Pipeline
Agent Builder through Agent Development Kit
Front End Interface

Inspiration

Finding information about the components we need and the risk for supply them can be complicated. Semiconductor supply chains have become increasingly difficult to predict. AI infrastructure growth, memory allocation constraints, advanced packaging bottlenecks, and shifting manufacturing priorities can dramatically affect component availability and demand. While most tools focus on pricing or compatibility, very few help users understand why a component may become difficult to source in the future.

What it does

ChipPulse AI is a semiconductor demand intelligence agent powered by Gemini, MongoDB Atlas, and Agent Development Kit (ADK).

Users can analyze a semiconductor component such as DDR5 memory, CPUs, GPUs, HBM memory, or other hardware products and receive:

A demand pressure score (0–100)
Risk classification (Stable, Watch, Elevated, High, Critical)
Supply availability assessment
Historical semiconductor events that resemble the current situation
Industry reports and market intelligence
News-based market signals
Actionable procurement recommendations
Agent reasoning and score explanations
Persistent analysis history

Unlike traditional search or recommendation systems, ChipPulse combines historical memory, live intelligence, and AI reasoning to explain why a component is becoming risky and what users should do about it.

How we start built it

Frontend

React + TypeScript dashboard
Interactive demand intelligence interface
Agent activity timeline
Historical retrieval visualization
Score breakdown and explainability panels
Deploy to Cloudflare Pages for simplicity

AI Layer

Gemini for reasoning and analysis generation
Gemini Embeddings for semantic understanding
Agent Development Kit (ADK) for multi-tool orchestration

MongoDB Atlas

MongoDB serves as ChipPulse's persistent memory layer.

Collections include:

historical_events
industry_reports
news_articles
analyses

We generate embeddings for all intelligence sources and store them in MongoDB Atlas.

Atlas Vector Search

MongoDB Atlas Vector Search powers semantic retrieval of:

Historical semiconductor disruptions
Industry reports
Market news articles

When a user analyzes a component, the agent performs semantic retrieval to find similar historical situations before generating a forecast.

Agent Architecture

ChipPulse uses a multi-tool agent architecture:

Intelligence Tools

Retrieve historical events
Retrieve industry reports
Retrieve market news
Evaluate component demand
Generate recommendations

MongoDB MCP Tools

Collection inspection
Aggregation queries
Memory analytics
Analysis history exploration

The agent retrieves relevant memory, reasons over the evidence, generates an assessment, and stores the result for future analysis.

Challenges we ran into

The first biggest challenges was building a deterministic grading system which avoiding generic AI responses, allowing vectors of news reports and other market factors to be taken into account for risk of supply and demand. Secondly, it was my first time building with the MongoDB MCP Server and the Google Vertex AI Agent, so there's some problems we had to resolved when implement these technologies.

Accomplishments that we're proud of

Built a working multi-tool AI agent using Gemini in Google AI Studio and Agent Development Kit
Implemented MongoDB Atlas Vector Search across multiple collections and utilizing the MCP toolbox for Agentic Workflow
Built a retrieval-based reasoning system grounded in historical events and market signals
Implemented agent activity tracking and retrieval provenance

What we learned

The advancement that Agentic Systems can provide when they use semantic, historical memory data for analysis.
The effectiveness of utilizing MCP toolbox to connect with Agent for further expansions of sources that can be added and controlled in the database