# AgenticEFI

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Arman Sahakyan](https://cerebralvalley.ai/u/brainsondemand)
- **GitHub:** https://github.com/nesall/agenticefi
- **Demo video:** https://youtu.be/DhirNIFxq4Y
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/425

AgenticEFI — Evidence-Driven Engine Calibration with Meta Muse Spark

For grassroots motorsports enthusiasts, including members of the Sports Car Club of America (SCCA), reliable engine calibration is essential for performance, consistency, and engine reliability. Yet amateur racers, track-day drivers, and independent tuners often face the same challenge: extracting trustworthy tuning information from thousands of noisy engine datalog samples.

Cold starts, throttle transients, fuel cut, and unstable operating conditions can distort air-fuel ratio (AFR) measurements. Naively averaging these samples can lead to misleading fuel-map corrections.

AgenticEFI addresses this problem by combining Meta Muse Spark with deterministic analytical tools to transform raw engine telemetry into conservative, evidence-backed Volumetric Efficiency (VE) tuning recommendations.

The application screens unreliable samples, analyzes AFR errors across engine operating regions, proposes bounded VE corrections, evaluates conditional improvements using held-out data, and provides traceable evidence for every proposed adjustment.

In the demonstration, AgenticEFI analyzes 56,284 real telemetry samples from a Mazda RX-7. Of 256 VE table cells, only five contain sufficient evidence for fitting, and just four correction proposals pass the acceptance criteria. This illustrates a fundamental principle: knowing when not to recommend a change is as important as identifying a useful correction.

Designed for the broader grassroots motorsports and DIY EFI community, AgenticEFI makes advanced datalog investigation more accessible while preserving engineering transparency.

All recommendations remain advisory. Predicted improvements are model-based estimates, not experimentally verified engine performance, and the application never automatically modifies an ECU.

Built with SvelteKit, TypeScript, Node.js, and Meta Muse Spark. The public repository includes a credential-free recorded investigation for independent evaluation.

## More from Meta Global AI Developer Hackathon

- [Team Phoenix](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/422)
- [RealityCheck](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/423)
- [Aathil](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/424)
- [Jean de Dieu Gasangwa](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/426)
- [The](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/427)
- [asaf meizner](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/428)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/425. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt