Every BivectorAI pilot starts with a written Scope of Work: exact inputs, expected outputs, success gates, non-claims, and data boundary definition. Your technical team reviews and approves it before any work begins. The output of the pilot is itself a hash-sealed BivectorAI evidence artifact.
Mỗi pilot bắt đầu bằng SOW bằng văn bản: đầu vào chính xác, đầu ra kỳ vọng, cổng thành công, phi-tuyên bố và định nghĩa biên dữ liệu. Đội kỹ thuật của bạn review và phê duyệt trước khi bất kỳ công việc nào bắt đầu.
A pilot is a controlled entry point — not a full exposure of the core. Customers evaluate measurable value without accessing protected internals.
The geometric physics framework, candidate-generation logic, and internal algorithms never leave BivectorAI's boundary during a pilot.
All customer-provided inputs stay inside the agreed data boundary. Air-gapped deployment available for defense, sovereign, and regulated customers.
Every pilot has a written Scope of Work defining inputs, outputs, success gates, non-claims, and data boundary — reviewed and approved before day one.
Every evidence artifact is SHA-256 sealed with provenance lineage and reproducibility instructions. The pilot output is itself auditable.
If your domain is not listed, contact us. The registry covers 64 domains. Each proposed pilot must first pass a suitability, evidence, and data-boundary review.
Engineering teams using AI coding assistants — GitHub Copilot, Cursor, Devin — where AI-generated pull requests touch protected paths, dependencies, or security-critical configurations without reliable review.
git diff · changed files · repository state · protected path list · policy file · evidence requirements. Air-gapped — no source code leaves your boundary.
PASS / REVIEW / BLOCK per change · JSON + Markdown evidence report · SHA-256 manifest · protected-path findings · dependency-drift findings · evidence-gap summary.
Every AI-generated change to your codebase exits the pilot with a hash-sealed evidence record your reviewer can challenge, override, or approve.
Materials research teams, semiconductor labs, battery manufacturers, and quantum hardware programs facing large candidate spaces before committing $50K–$500K per synthesis attempt.
Candidate structures · material class · target properties · constraint parameters · existing experimental baselines (if available).
Ranked candidate index · stability classification · provenance records · evidence gap map · expert-review handoff pack · SHA-256 manifest. 12 material groups covered.
96/96 quantum defect candidates converged in internal pipeline. Top score 97.59/100. Zero HPC or cloud required at screening stage.
Drug discovery teams, biotech startups, pharma R&D groups, and CROs committing $2M–$20M wet-lab budget — who need a structured, provenance-bound evidence package before the spend.
Molecular structures · target proteins · therapeutic area · existing assay data (if available) · CRO requirements.
Candidate identity controls · target mapping · docking evidence · ADMET/tox proxy · off-target proxy · novelty boundary · CRO-ready dossier · SHA-256 manifest. 49 protein targets covered.
253 preregistered candidate packages and 325 CRO-style evidence dossiers already prepared. Pilot can reference existing pipeline or screen new candidates.
Quantum software teams, regulated operators migrating to post-quantum cryptography, and organizations spending $10K–$100K per hardware session without workload provenance records.
QASM circuits · QEC artifacts · quantum workload definitions · cryptographic inventory · migration timeline requirements.
Artifact inspection report · bounded runtime evidence · failure fixture documentation · reproducibility receipts · PQC migration evidence pack · NDAA §889 allowlist enforcement records.
Air-gapped available. No quantum hardware required at evidence stage. Customer-controlled data boundary enforced throughout.
Research groups submitting beamtime proposals at GSI/FAIR, CERN, J-PARC, NSLS-II — where one failed or misconfigured session costs $50K–$500K and 12–24 months of program delay.
Experiment configuration · beamtime request records · computational lineage · event provenance files · energy budget data · milestone documentation.
Beamtime readiness assessment · configuration evidence pack · event provenance records · computational lineage · energy budget audit · milestone review gate · SHA-256 manifest.
GSI/FAIR (APPA, CBM, NUSTAR, PANDA) · CERN · J-PARC · NSLS-II · general HPC cluster evidence packages.
Robotics teams, industrial automation OEMs, counter-UAS operators, and GPS-denied navigation programs — where unverified sensor-to-action decisions carry $1M–$50M liability per incident.
Robot telemetry logs · sensor streams (radar, lidar, thermal, camera) · simulation replay data · action logs · policy candidates.
Sensor-state stability records · CUAS fusion evidence · GPS-denied navigation records · telemetry audit trail · action-boundary documentation · human-approval gate package.
Industrial robot arms · cobots · humanoid locomotion · counter-UAS · GPS-denied SAR navigation · ADAS · radar/lidar/thermal sensor fusion.
Grid operators, datacenter infrastructure teams, and OT/ICS/SCADA operators — where undetected instabilities cost $1M–$10M per hour and untraceable configuration changes create unlimited liability.
Grid state snapshots · configuration change records · cooling and power-flow data · OT/ICS event logs · infrastructure topology files.
Grid stability evidence at 5-second resolution · cooling and power-flow topology · infrastructure state records · causal incident reconstruction · change evidence trail · air-gapped deployment option.
Power grids · renewable integration · datacenter cooling · OT · ICS · SCADA · NERC BAL-003 / TPL-001 gate awareness.
Security teams, software vendors, and critical infrastructure operators — where a supply chain breach takes $100M+ and 6–18 months to scope and remediate without provenance records.
Software configuration records · dependency trees · change history · incident logs · OT/ICS/SCADA topology (air-gapped delivery supported).
Configuration lineage records · dependency provenance · causal incident reconstruction · policy-to-evidence mapping · CMMC Level 1–3 evidence records · air-gapped delivery.
CMMC Level 1/2/3 · DFARS clause mapping · NDAA §889 allowlist enforcement · supply chain SPOF detection.
Defense contractors, mission AI developers, and sovereign operators — where CMMC non-compliance risks losing all DoD contracts, and unaudited AI decisions carry mission and legal liability.
Mission data records · AI system outputs · C2 configuration logs · logistics data · compliance documentation. All data stays inside customer air-gapped boundary.
Mission-data provenance · hallucination firewall records · CMMC Level 1/2/3 evidence · DFARS clause mapping · NDAA §889 enforcement · PQC transition evidence · chain-of-custody records.
Air-gapped only. Customer-controlled cryptographic keys. No data leaves customer perimeter at any stage of the pilot.
Banks, fintech, cloud platforms, software vendors, and regulated infrastructure operators preparing for NIST PQC standards — who need a structured crypto inventory and migration evidence pack before board-level exposure.
Cryptographic inventory records · system architecture · current algorithm usage · compliance timeline requirements · regulatory context (DORA, NIS2, FIPS 140-3).
Crypto exposure review · migration roadmap evidence · board-ready summary · NIST PQC alignment records · regulatory flag documentation · SHA-256 sealed transition pack.
NIST PQC (ML-KEM, ML-DSA, SLH-DSA) · FIPS 140-3 · DORA · NIS2 · CHIPS Act · EU CRMA regulatory flag automation.
Financial institutions, sovereign wealth funds, and regulated infrastructure operators — where missing a Lehman-scale contagion signal carries $1B–$100B+ exposure and regulator sanctions.
Portfolio exposure matrices · interbank linkage records · supply chain concentration data · energy grid dependency maps · trade flow data. Privacy-aware — data stays inside customer boundary.
Systemic contagion evidence · supply chain SPOF detection · LCR hard-block gate · NERC BAL-003/TPL-001 grid gate · CHIPS Act / EU CRMA regulatory flags · regulator-readable evidence packs.
Basel III LCR unconditional gate · HHI SPOF threshold · Duck curve CAISO 2016 · interbank contagion topology. 227 tests passing in internal suite.
CFD teams, fusion plasma programs, structural engineering groups, and seismic hazard operators — where a single mesh inversion, plasma disruption, or seismic event costs $1M–$10M+ per occurrence.
Mesh files · solver configuration · plasma state snapshots · structural fatigue records · seismic monitoring data · induced seismicity logs.
Zero inverted mesh elements — guaranteed by construction · 3 unconditional plasma disruption hard gates · UK TLP GREEN/YELLOW/RED compliance · crack initiation onset detection · SHA-256 certificate per run.
43× mesh quality improvement in internal benchmark. 232 tests passing across 8 modules. No HPC required. TRL 1–5 across module range.
Medical research teams, academic hospitals, biotech R&D groups — who need reproducible, provenance-bound multi-signal cascade screening before clinical budget is committed. Research use only.
Anonymized patient records (no patient identifiers) · signal timeseries · lab values · imaging-derived data · biomarker panels. Anonymized case ID enforced throughout.
Multi-signal cascade geometry evidence · Bazett QTc with AHA/ACC 2009 gates · Sepsis-3 lactate hard block · CAR-T CRS hypothesis · amyloid onset screening · SHA-256 research certificate.
Cardiac · Sepsis/ICU · CAR-T · Amyloid · Immune · Autoimmune · Microbiome. 10 modules · 61 NOT_A_MEDICAL_DEVICE enforcement points in source. TRL 1–4.
Climate research groups, sovereign risk analysts, infrastructure planners, and reinsurance teams — who need geometric evidence of Earth-system cascade proximity before policy or investment decisions.
Climate state records · AMOC monitoring data · seismic monitoring logs · induced seismicity records · carbon flux data · Arctic/Amazon observational proxies.
AMOC bifurcation proximity evidence · Arctic albedo feedback gate · Amazon-permafrost compound cascade screening · UK TLP seismic GREEN/YELLOW/RED · seismic precursor detection · SHA-256 certificate per assessment.
7 Earth-system modules · 195 tests passing · SSP5-8.5 vs SSP1-1.9 scoring · seismic precursor 1,263× background detection. TRL 1–4. No operational forecast claimed.
Neuroscience labs, medical imaging research groups, and academic medical centers — where non-reproducible results invalidate years of downstream analysis and waste $500K–$5M in grant funding.
EEG · fMRI · Structural MRI · Protein system records · Pathology imaging · Genomic regulatory data. All data anonymized — no patient identifiers.
Research artifact indexing · subject-level evidence records · reproducibility receipts · imaging-derived data assessment · expert review handoff · provenance chain documentation · SHA-256 certificate.
Every result has a clear, challengeable provenance chain from raw data to finding. Designed for peer review, grant reporting, and replication by independent labs.
Media platforms, legal research teams, distributed systems operators, and database infrastructure teams — who need structural integrity verification without content analysis or viewpoint judgment.
Publication metadata · legal citation networks · distributed system operation logs · database transaction records · content provenance timestamps.
Temporal impossibility flags (no content read) · legal citation cycle detection (zero false positives on linear chains) · CAP theorem quantification · DB financial hard gates · equal treatment verification.
7 modules · 79 tests · Temporal integrity · Legal topology · CAP theorem · DB consistency · Echo chamber structural analysis. TRL 2–5. Free speech disclaimer enforced.
The platform covers 64 registered domains. If your use case spans multiple pilots or falls outside the 16 listed, contact us with your domain description. We will scope a custom pilot SOW within 48 hours.
Request Custom Pilot SOW →Choose domainSelect from 16 listed pilots or describe your use case.
Receive SOW in 48hExact inputs, outputs, success gates, non-claims, data boundary definition.
Your team reviews SOWTechnical team approves or requests changes before any work begins.
Controlled evaluationPilot runs within agreed data boundary. Air-gapped if required.
Receive evidence artifactHash-sealed PASS/REVIEW/BLOCK report with provenance and reproducibility receipt.
Expert reviewYour technical team challenges, verifies, and audits the output.
Go / No-Go decisionWritten go/no-go result. Clear next commercial step if go.
Founder: Dam Van Vi · damvanvi@gmail.com · ORCID 0009-0006-6039-0989