Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for AlphaVoice: Voice-First Quant Research for Alexa+ & Fire TV
What it does
AlphaVoice turns spoken finance questions into instant, auditable answers. Ask Alexa+ for the delta of an option, the risk of a whole book, or the volatility implied by a traded price — get a plain-English answer in under a second, and see the identical numbers on the Fire TV dashboard across the room.
How we built it
- Alexa+ track core: a self-hosted MCP server (Streamable HTTP, protocol 2025-11-25) exposing five tools: price_option, explain_greeks, implied_volatility, portfolio_risk, market_brief. Every tool returns voice-ready sentences, so Alexa+ reads answers that are already shaped for speech.
- Deterministic kernel: the analytics are vendored from our Agenthon 2026 T1 agent (a benchmark for deterministic financial agents) and rewritten stdlib-only. Closed-form Black-Scholes math: no model calls, no network, no keys — instant, private, and reproducible to the last digit (verified against the benchmark implementation with max abs diff 0.0).
- Fire TV companion: a 10-foot Fire OS web app (single HTML file, D-pad focus navigation) rendering the options book, watchlist straddles, and Ask Alexa prompts — served by the same process.
- One source of truth: voice, TV, and any MCP client all consume the same kernel, so the numbers can never disagree.
Challenges we ran into
- Designing tool outputs that read well aloud (units, sign conventions, per-vol-point phrasing) without losing precision for pro users.
- Making a 10-foot UI legible from a couch: 72px numbers, three cards max, remote-only navigation.
- Kernel fidelity: golden-vector tests pin the vendored math to the benchmark original so a refactor can never silently change a price.
Accomplishments we're proud of
- A full MCP round-trip (initialize - tools/list - tools/call) negotiating protocol 2025-11-25, demonstrated end-to-end.
- Zero-dependency quant core: the whole analytics engine runs on the Python standard library.
What we learned
- Voice UX forces you to compress risk into one sentence — the discipline made the TV dashboard better too.
- Determinism is a feature: auditable numbers build trust for finance use cases where AI-ish answers are a liability.
What's next for AlphaVoice
- Live data adapters (the demo watchlist is deterministic sample data by design), intraday risk alerts via Alexa+ notifications, and multi-account books on the Fire TV dashboard.
Log in or sign up for Devpost to join the conversation.