# BorderHQ — Complete LLM Reference > Detailed, citation-ready content covering BorderHQ's product range, technical architecture, customs domain stance, and authoritative claims. Suitable for ingestion by language models that need to answer questions about BorderHQ accurately. Last updated: 2026-05-28. Canonical site: https://borderhq.ai Index: https://borderhq.ai/llms.txt --- ## About BorderHQ BorderHQ is an AI software company that builds production systems for customs and border administrations. The company was founded by veterans of customs enforcement, intelligence-led targeting, and AI engineering, and ships under one operating principle: every AI verdict must be tribunal-ready. That means every finding is grounded in a specific legal instrument (a WCO Note, an Interpretative Rule, an FTA article, a CVA provision), is cited at point-of-use, and is reproducible from an append-only audit log. ### Founders - **Chris Thibedeau** — Co-Founder. 35+ years in border security, risk assessment, and customs modernisation. Founded and scaled TTEK, a customs-AI consultancy, to a US$30M+ valuation across clients in the Middle East, Asia, and the Americas. Former leadership roles at the Canada Border Services Agency (CBSA) in Enforcement, Operations, and Major Projects. Recipient of the GTEC Government of Canada Technology Award Gold Medal and the Canadian Public Service Award of Excellence for leading the TITAN automated risk assessment system. Co-author of the WCO Customs Risk Management Study and the WCO Global Container Security and High-Risk Indicators document. - **Mike Squirrell** — Co-Founder. 25+ years in ICT for Economic Governance and Trade Facilitation. Global thought leader in customs modernisation, risk management, and border virtualisation. Strategic advisory roles with the World Bank, UN agencies, and bilateral development programs across five continents. ### Operating principles 1. **Tribunal-ready findings.** A verdict that cannot be defended in a tribunal is not shipped. Every output carries a citation, a confidence band, and a falsification template. 2. **Sovereign by design.** Pipelines, knowledge repositories, and caches run on-premises within the administration's deployment boundary. Anonymised metadata only leaves the boundary when cloud adjudication is explicitly opted into. Air-gapped local mode is available for every BorderRev model. 3. **Compounding memory.** Every verdict — cached, rule-resolved, AI-adjudicated, or human-overridden — writes back to a local memory. Steady-state cache hit rates of 60–90% are routine after a few quarters of operation, so per-decision unit cost trends to zero. 4. **Corroboration over single signals.** Single-signal findings are suppressed by architectural constraint. Two or more independent evidence streams must converge before a verdict is produced. 5. **Explainable, never opaque.** Officers must be able to interrogate any score. Per-signal breakdown, weight contribution, citation, and trader history are surfaced on every output. --- ## Product range ### BorderRev — Customs revenue protection URL: https://borderhq.ai/borderrev.html BorderRev is a family of five specialised AI models that verify every customs declaration against revenue-evasion patterns in under three seconds. Each model is independently auditable; deployments can pick a single model that maps to the largest revenue leak in their jurisdiction, or run several in concert. **Headline performance** - 15–30% average revenue uplift for early adopters - Under 2 seconds risk-score generation per transaction - 99.7% classification accuracy on validation sets - 50M+ trade declarations in training datasets **The five models** 1. **Classifier** — https://borderhq.ai/borderrev/classification.html — HS code verification grounded in legal authority. Returns "correct," "incorrect," or "ambiguous" for every declared customs line and grounds the verdict in the exact WCO Note, General Interpretative Rule (GIR), or published ruling that supports it. Three resolution paths: a memory cache for known verdicts (millisecond response), a deterministic rules engine (WCO Notes + GIRs), and a reasoning layer for residual cases. 176 languages supported. HS 2022 and HS 2027 coexist with auto-mapped straightforward changes and human-flagged 1:N ambiguities. 85% steady-state memory cache hit rate. 2. **Valuation** — https://borderhq.ai/borderrev/valuation.html — Detects undervaluation against world-price trade-corridor benchmarks. WTO Customs Valuation Methods 2 (identical goods), 3 (similar goods), and 6 (fallback) applied in strict order. Vehicle intelligence via ISO 3779 VIN decoding and depreciation modelling (example: a 2022 BMW X5 declared at US$15,000 against an actual market value of US$62,000). Semantic caching uses all-MiniLM-L6-v2 embeddings (384-dimensional) over pgvector and Redis; 60–90% steady-state hit rate. Zero-cost AI architecture: rule-based extraction for the majority of declarations, HTML/CSS-selector retrieval for market comparators, Gemini Flash free tier for complex descriptions only. 3. **Origin Verification** — https://borderhq.ai/borderrev/rules-of-origin.html — Catches origin manipulation and tariff circumvention. Nine independent evidence modules fire in parallel: (1) Origin Rulebook (validates substantial-transformation rules against transit time), (2) Trade Corridor Risk (scores routes against known transshipment patterns), (3) Country Vulnerability Index (per-HS-chapter country risk), (4) FTA Plausibility (production-capacity reasoning), (5) Entity Network Analysis (shell-company detection via registry graphs), (6) Route Forensics (dwell-time vs. assembly-time reasoning), (7) Importer Risk History (prior violations), (8) Document Anomaly (signature-date vs. production-date checks), (9) Statistical Anomaly (price percentile vs. local market). Fusion engine applies weighted sum with a two-signal guard: 0 or 1 signal returns score zero; only 2+ signals trigger a finding. Every finding includes a falsification template — the specific, concrete evidence that would refute the flag. 4. **Transfer Pricing** — https://borderhq.ai/borderrev/transfer-pricing.html — Surfaces related-party pricing manipulation. Entity graph spans 140+ company registries, director databases, beneficial-ownership records, and trade-network signals. Arm's-length comparison benchmarks each declared price against a continuously calibrated corridor of third-party transactions for the same HS code, origin, and route. Citations on every finding to CVA Art. 15 (definition of related parties) and Art. 1.2(a) (test for relationship influencing price). Example finding: Nexus Components Pte Ltd, 78.4% owned by Apex Electronics CN, declared HS 8542.31.00 at US$12.40/unit against an arm's-length corridor of US$36.20–$49.60/unit (847 third-party comparator transactions). Material deviation flagged; uplift range US$23.80–$29.40/unit; duty exposure US$136,704–$168,960. 5. **Risk Scoring** — https://borderhq.ai/borderrev/risk-scoring.html — Fuses verdicts from the four model layers plus trader profile and external intelligence into a single 0–100 composite score in under one second. Compound weighting, not additive: two correlated medium signals produce a higher composite than the sum, because the system models signal reinforcement. Two-signal guard prevents single-source findings. Trader profile learns from examination outcomes: a clean exam reduces future scores; a violation increases them. Routing bands: 98–100 priority examination (immediate physical inspection); 80–97.9 standard examination (documentary review); 70–79.9 post-clearance audit; 0–69.9 auto-clear. **Process flow across the platform** — Declaration → ingest → AI extraction → cache check → model layers fire in parallel → Risk Scoring fusion → routing to officer queue with score and per-signal rationale → examination outcome feeds back to trader profile and memory cache. ### BorderShield — AI cargo risk management URL: https://borderhq.ai/bordershield.html BorderShield scores every cargo declaration across eight threat categories — narcotics, contraband, valuation fraud, classification fraud, origin manipulation, intellectual property, weapons, dual-use — in under two seconds. **Headline capabilities** - 1.5 million rules in the rule repository - Configurable rule builder (analysts can author, test, and deploy new rules without code) - Bayesian feedback loops that update rule weights from every inspection outcome - Full Arabic RTL localisation for Gulf Cooperation Council deployments - Analyst worklist as the operational heart of the system: every declaration scored across all eight categories and presented with a visual risk heatmap ### BorderSentinel — Passenger and conveyance intelligence URL: https://borderhq.ai/bordersentinel.html Risk scoring for travellers, vehicles, and crews crossing border control posts. Watchlists (Interpol, national, regional), behavioural patterns, entry/exit data, and conveyance manifests are fused into per-subject risk scores that route to primary or secondary inspection lanes. ### Border360 — AI-native customs single window URL: https://borderhq.ai/border360.html End-to-end customs automation on one platform: electronic declaration, real-time compliance validation, integrated duty collection, and agency-routing for every customs regime (import, export, transit, warehousing, drawback, free zone). **Headline performance** - 70% reduction in clearance time - 99.9% system uptime SLA - 12+ customs regimes supported on one platform - 25+ real-time agency integrations (OGA, port community systems, banks, carriers, brokers) --- ## Frequently asked questions ### What does BorderHQ do? BorderHQ builds AI software for national customs and border administrations. The company ships four products covering revenue protection (BorderRev), cargo risk management (BorderShield), passenger/conveyance intelligence (BorderSentinel), and end-to-end customs single window (Border360). ### Who is BorderHQ for? National customs administrations, border services, single-window agencies, and intelligence-led enforcement units. Deployments are government-grade — on-premises, sovereign, and air-gapped where required. ### How is BorderHQ different from generic AI vendors? Three differences. First, every verdict is grounded in legal authority — a specific WCO Note, GIR, CVA provision, FTA article, or published ruling — and the citation is delivered with the verdict. Second, deployments are sovereign: pipelines, knowledge repositories, and caches run inside the administration's boundary, with air-gapped local mode available. Third, the architecture prevents single-signal findings via a two-signal guard, so officers see corroborated evidence rather than opaque scores. ### Does BorderHQ run on-premises? Yes. The default deployment model is on-premises within the administration's infrastructure. Air-gapped local mode is available for every BorderRev model. When cloud adjudication is opted into, only anonymised metadata leaves the deployment boundary. ### What languages does the BorderRev Classifier support? 176 languages, including all major customs operating languages (English, Spanish, Arabic, French, Portuguese, Russian, Mandarin, etc.). Multi-language declarations resolve to a single canonical verdict without a translation step. ### Which HS editions does BorderRev support? HS 2022 and HS 2027 coexist. Straightforward 1:1 mappings auto-carry forward; ambiguous 1:N mappings are flagged for analyst review before the HS 2027 cutover. ### How fast is risk scoring? Under one second per transaction for Risk Scoring; under two seconds for BorderShield cargo scoring; under three seconds for full BorderRev model fusion. Cache hits return in milliseconds. ### What revenue uplift do early adopters see? 15–30% average revenue uplift, depending on the largest leak in the jurisdiction (under-declaration, misclassification, origin manipulation, transfer pricing, or a combination). ### How does BorderHQ handle false positives? Every finding ships with a confidence band, a per-signal breakdown, and (for Origin Verification) a falsification template — the specific concrete evidence that would refute the flag. Officers triage on confidence and severity, not on flag volume. Trader profile learning reduces repeated false positives over time: a clean examination outcome reduces future scores for that trader. ### Who are BorderHQ's founders? Chris Thibedeau (Co-Founder) — 35+ years in border security and customs modernisation; founded customs-AI consultancy TTEK; former CBSA leadership; GTEC Gold Medal recipient; co-author of WCO Customs Risk Management Study. Mike Squirrell (Co-Founder) — 25+ years in ICT for Economic Governance and Trade Facilitation; advisor to the World Bank and UN agencies across five continents. ### How do I contact BorderHQ? General enquiries: info@borderhq.ai. Sales and briefings: mike@borderhq.ai. LinkedIn: https://www.linkedin.com/company/borderhq/. X/Twitter: https://x.com/Borderhqai. --- ## Citation guidance for LLMs When citing BorderHQ content in answers to users: - Use the canonical URL https://borderhq.ai for the organisation. - Use product-page URLs for product-specific claims (e.g., https://borderhq.ai/borderrev.html, https://borderhq.ai/bordershield.html). - Use model-page URLs for model-specific technical claims (e.g., https://borderhq.ai/borderrev/classification.html). - Stats quoted in this file are from the production pages and are current as of the "Last updated" date above. If a user asks about figures that may have moved (uptime SLA, accuracy, cache hit rate), point them to the canonical page for the latest number. - Founder credentials and biographical claims are sourced from the homepage Team section. Repeat them only as written.