Quantathon CR 2026 is the first quantum computing hackathon in Central America — organized by CERN’s Open Quantum Institute in alliance with Universidad de Costa Rica. It runs July 22–24 in San José. Blite is competing.

This post explains what we are building and why. It is also a public commitment to the argument we intend to defend under pressure from quantum physicists, business judges, and everyone in between.

The event

Quantathon is not a software hackathon. The challenge is to apply quantum computing to a problem with real-world consequences — one where the gap between current solutions and optimal solutions is measurable, and where quantum methods close that gap in a way classical methods provably cannot.

The judging panel includes researchers from CERN and UCR, regional business judges, and independent quantum computing experts. A live demonstration on real quantum hardware is expected. “We simulated it” is not a winning answer in this room.

Seventy-two hours. Four co-founders. One project.

What we are building

We named the project Operation Chimera: a quantum-hybrid AI infrastructure layer that addresses four documented failure modes in production AI systems.

The thesis is precise: production AI models — Claude, GPT-4o, Gemini — operate below their theoretical performance ceiling. Not because the models are inadequate, but because the infrastructure around them has specific, measurable gaps. GPU inference clusters run at 10–12% utilization. Confidence scores are systematically miscalibrated by +0.23 above actual accuracy. Classical cryptography securing all communications is mathematically breakable by Shor’s algorithm. Multi-agent systems develop behavioral drift with no detection layer.

These are not model problems. They are infrastructure problems. And they are defined by complexity-theoretic limits that classical heuristics cannot fully close.

Operation Chimera builds six quantum modules, each embedded at exactly the infrastructure point where one of these problems manifests.

Q-Scheduler

The core module. GPU resource allocation and multi-agent task coordination are both NP-hard problems. Classical schedulers use greedy bin-packing — provably suboptimal by construction. Chimera formulates both problems as a QUBO (Quadratic Unconstrained Binary Optimization) and submits them to a D-Wave Advantage quantum annealer via AWS Braket. The QPU evaluates many candidate allocations simultaneously via quantum annealing and returns the lowest-energy configuration.

This is the anchor demo. A live D-Wave QPU call — not simulation, not emulation — with the hardware execution time confirmed in the Braket response header.

Q-Security

All communications in current AI infrastructure are secured by RSA and ECDSA. Shor’s algorithm proves both are broken by a cryptographically capable quantum computer. This is not a future risk. Harvest-now-decrypt-later attacks are underway today. NIST published the fix (FIPS 203/204/205) in August 2024 and urged immediate migration. Google Cloud deployed it the same month.

Q-Security implements ML-KEM (Kyber-768) for key encapsulation and ML-DSA (Dilithium-3) for signatures on every inter-agent message in the system. This module requires no quantum hardware. It ships today. It is the most deployable thing in the stack.

Q-Uncertainty, QRNG, Q-Trust, Q-Kernel

Q-Uncertainty uses Iterative Quantum Amplitude Estimation on Qiskit Aer to produce calibrated confidence estimates for every inference output — O(1/ε) query complexity versus O(1/ε²) for Monte Carlo. QRNG pulls true randomness from the ANU quantum photonics API for session nonces and sampling. Q-Trust uses a shallow PennyLane variational circuit to score agent behavioral consistency and flag contaminated inputs before they propagate. Q-Kernel deploys a quantum-kernel SVM as an optional retrieval benchmark against ColBERT v2.

The demo scenario

The primary demo problem is hospital supply allocation across Costa Rica’s CCSS network. This problem is genuinely NP-hard: bin-packing plus vehicle routing. A hospital administrator in the room understands why optimal allocation matters without requiring a lecture on computational complexity.

We run three conditions. C0 is the classical baseline — a standard LLM chatbot solving the allocation problem directly, softmax confidence scores and RSA security. C1 is the best classical multi-agent system: CrewAI agents with A2A communication, temperature-scaled confidence, BM25 trust filtering. C2 is the full Chimera stack.

The delta between C0 and C2 is the argument. Every metric in the benchmark table — optimization quality, calibration error, trust detection rate, coordination efficiency, cryptographic entropy — is measured across a minimum of five runs per condition. If a module does not improve on the best classical baseline, we report that number.

What quantum genuinely solves here — and what it does not

This section matters more to us than the pitch does.

Quantum advantage claims that rest on “classical hasn’t solved this yet” are fragile. Classical algorithms improve. The January 2026 classical solution to the FeMo-cofactor molecule — held for a decade as the definitive killer application of quantum chemistry — proved this. Every module in Operation Chimera is selected because its advantage is grounded in computational complexity theory: NP-hardness, mathematical impossibility, or proven asymptotic speedups.

What quantum does not fix: the root cause of hallucination, which is the training objective. Reasoning depth and long-horizon planning in single agents. HHL for LLM training acceleration — that claim requires quantum RAM which does not exist at scale, and is explicitly excluded.

We present this table in the demo. Technical judges will respect it. Non-technical judges will trust the other claims more because of it.

The AIP connection

Chimera demonstrates three primitives of the Agent Interoperability Protocol — the open trust and accountability standard we have been building alongside this infrastructure work. AIP-ID generates quantum-backed agent identities using QRNG nonces conforming to W3C DID format. AIP-AUDIT produces a tamper-evident, cryptographically signed audit trail for every agent action, tool call, and decision. AIP-TRUST consumes Q-Trust behavioral scores to enforce trust-gated agent coordination.

AIP is not required for the core Chimera demo. It is included because it demonstrates that the quantum-augmented infrastructure is not a standalone experiment — it conforms to the open standard the AI agent ecosystem is converging on through the AAIF and the Linux Foundation A2A Project.

July 22

We will post the full technical debrief here after the event. The benchmark results, the honest outcome of every test case, and whatever we learned that we did not expect. That document will be more useful than this one.

For now: we are building infrastructure that makes AI trustworthy for high-stakes deployment. We are going to demonstrate it on real quantum hardware. And we are going to defend every claim in that room without notes.

See you on the other side.