DataCompass

Inspiration

Startup investors and founders often need to benchmark a company against competitors, analyze market position, and identify strategic opportunities. But doing this manually through thousands of profiles is time-consuming. We wanted to create an autonomous, intelligent agent that could perform this analysis in seconds using natural language—combining Google Cloud technologies with a multi-agent architecture.

What It Does

DataCompass is an AI-powered agent that allows users to input natural language descriptions of startups (e.g., "We're a Series B cloud computing company with $6M revenue") and returns deep insights, including:

Competitive benchmarking with similar companies
Vector-based similarity analysis using BigQuery ML
Real-time insights from a dataset of 50,000+ startup profiles
A conversational interface for follow-up questions (e.g., "How does my funding stage compare?")

How We Built It

We used:

Google Agent Development Kit (ADK) to build a multi-agent workflow
Vertex AI for generating text embeddings (text-embedding-004)
BigQuery for querying and joining startup metadata and vector similarities
BigQuery ML for VECTOR_SEARCH on company embeddings
A Java tool to generate and insert embeddings into BigQuery
A cleaned Kaggle Crunchbase dataset as our data foundation

The agent workflow consists of:

Analysis Agent – Processes user input and extracts attributes
Benchmarking Agent – Performs vector similarity search and generates insights

Challenges We Ran Into

Integrating BigQuery ML’s VECTOR_SEARCH into an automated Java pipeline
Generating and managing large-scale embeddings efficiently
Structuring multi-agent communication in ADK for both analysis and benchmarking
Parsing unstructured natural language input reliably for vector encoding

Accomplishments That We're Proud Of

Building a fully working multi-agent system using Google Cloud’s newest agent platform
Running real-time vector similarity search across 50,000+ companies with sub-second latency
Creating a user-friendly and powerful interface for startup analysis powered entirely by agents

What We Learned

How to design and orchestrate agent workflows using the Agent Development Kit
Best practices for embedding generation with Vertex AI
Efficient use of BigQuery ML's vector capabilities for real-world business analysis

What's Next for DataCompass

Crunchbase API integration for live, up-to-date data
Add more sub-agents, such as:
- Competitor tracking via web scraping or APIs
- Sales call summarization via transcription + NLP
- Trend detection and market forecasting
Expand the UI for more user-friendly dashboards and collaboration tools