ExecuTV — AI Executive Workstation for Fire TV

Elevator Pitch

ExecuTV turns Fire TV into a voice-driven executive workstation. Using Alexa+ MCP and AWS, leaders track KPIs, check Ring alerts, and approve high-stakes decisions hands-free.

💡 Inspiration

I've always been fascinated by the idea of computers that are truly hands-free. In the past, I tried building a hands-free driving assistant that gave drivers location-based reminders as they passed specific places, logged observations on the road, and surfaced those reminders once they got home.

When the Build, Ship, Shape: Amazon Developer Hackathon 2026 was announced, I planned to enter the Bee (Wearable AI) track. However, that track requires live data recorded on a physical Bee device or an Apple Watch running Bee software, and neither was available in South Africa.

Rather than let hardware stop the project, I pivoted to Fire TV and Alexa+, which the rules allow to be built with simulated Alexa+ web agents and the Fire TV / Vega OS simulator. I realized the living room is the ultimate hands-free frontier: a place where busy executives can spend time with family on the couch while staying connected to the business. They can monitor KPIs, inspect site alerts, and approve decisions without reaching for a laptop or phone.

🚀 What It Does

ExecuTV turns the living room TV into a voice-first executive workstation. Executives speak naturally, and ExecuTV responds with a spoken summary and a synchronized big-screen dashboard.

Morning Briefings: Say "Alexa, give me today's brief" to hear a summary of company revenue, key margin metrics, and pending approvals while high-contrast KPI cards appear on screen.
Root-Cause Variance Analysis: Ask "Why is margin down in Region B?" to see animated trend charts while the AI explains the underlying drivers, such as a supplier price shock and logistics rerouting.
Strategic Challenge Gate: When an executive requests a high-risk action, such as a $1.2M emergency procurement, ExecuTV doesn't simply comply. Code-enforced policy guardrails trigger a one-turn challenge from a domain persona (the CFO). The TV shows a Strategic Audit Card with a risk scorecard, the riskiest underlying assumption, and a 7-day action plan before anything is executed.
Operational Workflows & Ring Alerts: Inspect blocked workflows, such as a delayed month-end close, and view Ring camera freight-delivery snapshots (simulated for the demo) directly on the TV.

🛠️ How I Built It

ExecuTV uses a decoupled cloud architecture:

Alexa+ MCP Server (apps/ingress_gateway): A self-hosted Model Context Protocol server over Streamable HTTP (spec 2025-11-25), with Bearer token authentication.
Core Orchestration Engine (apps/core_engine): A FastAPI service that routes intents, applies standing playbook rules, and produces a DualPayload: a voice_summary for Alexa and a visual_state_model for the TV.
Voice-Visual Sync: Visual updates are pushed to the React TV interface over a persistent Server-Sent Events channel (/events/ticker), so each tool call from Alexa directly drives which card is highlighted on screen. The interface runs in [the Fire TV / Vega OS simulator / a TV browser — confirm].
Data & Scenario Seed: PostgreSQL (packages/core_data_layer) holds a fully linked scenario for a fictional company, AcroCorp Global Technologies, so financials, supplier price shocks, logistics bottlenecks, and simulated Ring camera feeds stay consistent across every query.
AWS Infrastructure: ECS Fargate behind an Application Load Balancer (idle timeout raised for long-lived SSE connections), RDS PostgreSQL, S3 + CloudFront for static hosting, and the Amazon Bedrock Converse API running [model name].

🚧 Challenges I Ran Into

Building without physical hardware: Without access to Bee devices, I had to translate hands-free productivity from a wearable to a 10-foot living room display, working entirely in simulators.
Syncing voice and screen across separate clients: Tool calls arrive from a remote Alexa+ agent, but the visuals live on a separate TV client. I solved this by linking sessions to a server-push SSE stream.
Keeping the demo consistent for every judge: To stop one reviewer's actions from changing the scenario for the next, I added a token-protected reset endpoint (POST /admin/reset-scenario).

🏆 Accomplishments I'm Proud Of

Spec-compliant remote MCP server: A working, self-hosted Streamable HTTP MCP server (spec 2025-11-25), tested with Claude Desktop and MCP Inspector.
Deterministic guardrails plus generative reasoning: Code-enforced risk tiers (automatic approval for low-risk actions, a one-turn challenge for high-risk ones) combined with LLM-generated explanations.
Tight voice-visual sync: Sub-second card focus updates on the TV, timed with the spoken response.

📚 What I Learned

How to run remote MCP tool routing over Streamable HTTP instead of local stdio connections.
How to design 10-foot UIs for lean-back viewing: high contrast, large type, and clear remote-control focus states.
How to balance generative AI reasoning with strict, deterministic business rules in an enterprise setting.

🔮 What's Next for ExecuTV

Multi-executive collaboration: Voice profile recognition so several executives can use it together in board-style sessions.
Live ERP integrations: Connecting the core engine to SAP and Salesforce alongside the AcroCorp demo dataset.
Native app packaging: Packaging the TV interface as a native Fire OS / Vega OS app for store distribution.

🏷️ Built With

alexa fire-tv ring aws amazon-bedrock mcp model-context-protocol fastapi postgresql react vite python typescript docker terraform