Devpost Submission: Lighthouse 3 - Strategic Intelligence Advisor

Inspiration

Executives today are drowning in "AI Noise." Between the endless stream of press releases, hallucinated LinkedIn hype, and dense technical whitepapers, it has become nearly impossible for a C-Suite leader to answer the most critical question: "What does this actually mean for our ROI?" We built Lighthouse 3 to act as a digital "Chief of Staff." We wanted to create a beacon of clarity—a tool that doesn't just summarize the news, but filters it through a strategic lens, identifies non-obvious market connections, and provides an auditable logic path for its conclusions.

What it does

Lighthouse 3 is an autonomous "High-Reasoning" research agent that monitors the global market. It performs deep-dive "Intelligence Hunts" to produce professional-grade Strategic Briefings rendered through a minimalist web portal.

  • Strategic Filtering: It ignores general hype to focus on four pillars: ROI & Business Impact, Competitive Intelligence, Regulatory Risk, and Hidden Connections.
  • Verified Synthesis: It uses Google Search Grounding to anchor every claim in real-world, real-time sources.
  • Logic Auditing: Every report features a "Thought Signature"—a transparent log of the agent's internal reasoning—so executives can trust how the AI arrived at its strategic advice.

How we built it

Lighthouse 3 is built on a "Lean-Executive" architecture designed for speed and reliability:

  • Core Brain: Powered by Gemini 3 Pro Preview, utilizing the thinking_level=HIGH configuration to enable multi-step reasoning before output generation.
  • Grounding Engine: Integrated with the Google Search Tool via the google-genai SDK to ensure real-time accuracy.
  • Portal Backend: A Python/Flask server that dynamically renders Markdown reports into a clean, "Boardroom-Ready" interface.
  • Cloud Infrastructure: Containerized with Docker and deployed on Google Cloud Run. This serverless approach ensures the "Intelligence Portal" is highly available and scales with executive demand.

Challenges we ran into

The most significant hurdle was Deployment Data Persistence. We faced a classic "missing file" mystery where our briefings weren't appearing in the cloud. We overcame this by mastering Cloud Run's deployment lifecycle, implementing absolute pathing within the container, and engineering a precision .gcloudignore file to filter out thousands of virtual environment files while "force-allowing" our mission-critical Markdown reports. This taught us that in the cloud, infrastructure is just as important as the model logic.

Accomplishments that we're proud of

  • The "Hidden Connections" Logic: We successfully enabled the agent to link disparate news items—like connecting infrastructure pivots to secondary energy market shifts—providing "second-order thinking" usually reserved for human analysts.
  • Production-Grade Deployment: We moved from a local script to a fully containerized cloud application with a dedicated Service URL, managed via the Google Cloud SDK.
  • Zero-UI Design Philosophy: We skipped the "chatbot" cliché. By focusing on a "Push-Intelligence" model, we created a tool that respects an executive's time by delivering answers, not conversation.

What we learned

We learned that transparency is the key to AI adoption in the C-Suite. An executive doesn't just want an answer; they want to see the work. Building the "Thought Signature" taught us that exposing the AI's reasoning process is more valuable than any complex dashboard. We also gained deep experience in Atomic Revisions on Cloud Run—learning how to push updates and reports to a live URL without disrupting service.

What's next for Lighthouse 3: Strategic Intelligence Advisor

  • Dynamic Persistence: Moving reports from the container filesystem to Google Cloud Storage (GCS) for instant, "no-deploy" briefing updates.
  • Enterprise Integration: Delivering briefings directly to secure executive Slack channels or encrypted email inboxes.
  • Long-Term Memory: Implementing a "Contextual History" so Lighthouse 3 can track how its own strategic predictions have played out over months, providing a feedback loop for its own reasoning.

Built With

Share this project:

Updates