Skip to Main Content

Antevorta

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Antevorta — Demo video

Antevorta is an intelligent, high-conviction decision stress-testing workbench built for high-stakes life and career choices ("Test your decisions before you live them"). Most major decisions fail not from bad intentions, but from uncalculated constraints: weekly hour overcommitments, rapid financial burn, or unverified assumptions. Today, generic AI tools act as sycophantic cheerleaders or hallucinate made-up success percentages (e.g. "You have an 85% chance of success!"). Antevorta enforces a strict architectural boundary: deterministic Python math calculates all weekly capacity, cashflow runway, and breakeven thresholds, while Meta's Muse models reason over tool-verified facts to surface conditional trade-offs, critical unknowns, and reversible experiments. Key features built during the hackathon: 1. Deterministic Feasibility & Runway Engine: Computes exact weekly headroom against the 168-hour ceiling, monthly runway buffers, and prerequisite verification. Missing values stay UNKNOWN, never zero. 2. Bounded Meta Agent Tool-Calling Loop: Muse Spark 1.3 autonomously invokes allow-listed local tools (assess_option, compare_options, list_unknowns, simulate_sensitivity) with self-healing critique recovery. 3. Zero-Hallucination Verifier: Deterministically inspects prose outputs to eliminate fabricated probabilities or unsupported numbers. 4. Adaptive Clarification Engine: Computes an intake Clarity Score (0–100) with 1-click assumption patches. 5. What-If Sensitivity Sandbox: Real-time slider recalculation with structured "What Changed & Why" diffs and stale narrative alerts. 6. Calibration Mirror (L4A): Implements a machine-verifiable prediction ledger evaluated with the Brier Skill Score (BSS = 1 - BS_twin / BS_generic). 7. Person Model (L3) & Topological Decision DAG (L6): Exponential claim decay and bottleneck analysis. 8. Multi-Model Meta Suite: Grounded Future Self dialogues (Muse Spark 1.3), spoken dilemma intake via Web Speech API + Muse Spark 1.3, and Day-in-the-Life visual scene generation (Muse Wonder 1.0 + Muse Image 1.0). 9. Archival Codex & Warm Vellum UI: Editorial design system with bearer-token privacy, GDPR export, and one-click account purge.

Team