--- title: "Manufacturing AI: predictive maintenance, vision, edge" description: "Production AI for manufacturers and supply-chain operators. Predictive maintenance, vision defect detection, OT/IT bridge, and edge inference that survives 24/7 line speed." source: "https://www.kensink.com/industries/manufacturing-ai/" canonical: "https://www.kensink.com/industries/manufacturing-ai/" --- ★ Industry vertical · Manufacturing AI 06 service items Edge + OT engagements INDUSTRY · MANUFACTURING AI · OT + SUPPLY CHAIN # AI for the factory floor that keeps shipping. Production AI for manufacturers, supply-chain operators, and industrial OT teams. Predictive maintenance that survives the drift, vision defect detection that hits line speed, OT/IT bridges that respect operations, and an eval cadence that does not break a 24/7 shift. [Start a project →](https://www.kensink.com/contact) [View case studies →](https://www.kensink.com/cases) Industry Manufacturing · OT · supply chain Environment ISA-95-aware · edge + cloud Uptime 24/7 · production-grade Stack Edge inference · auditable \[WHAT WE HEAR FROM PLANT MANAGERS AND OT CISOs\] ## Three pains every manufacturing-AI team hits. We have shipped against this shape with discrete manufacturers, process-industry operators, supply-chain platforms, and industrial robotics teams. The constraints repeat. The fix is not a fancier model. The fix is operations-discipline: edge inference, drift detection, OT-respecting integration, and an architecture that survives the next five-year capex cycle. PAIN · 01 01 / 03 ### Predictive maintenance models drift on the floor. Sensor calibration changes, supplier lots change, line conditions change. A model that was 92% accurate at install drops to 70% in six months and nobody notices until a critical asset fails. ↓ How we fix it, below. PAIN · 02 02 / 03 ### OT integration is a different beast. SCADA, PLCs, MES, historians, ISA-95 boundaries. Most AI vendors treat the factory like a SaaS app. OT engineers reject that on day one and the project never reaches the line. ↓ How we fix it, below. PAIN · 03 03 / 03 ### Architecture lives across five-year capex cycles. Industrial systems do not get rebuilt every two years. Edge hardware, model registry, OT bridge — those decisions get inherited by the next plant manager and the one after that. Wrong call costs a capex round, not a refactor. ↓ How we fix it, below. \[SIX SERVICE ITEMS · ONE TEAM\] ## Pick the manufacturing-AI problem. We'll bring the build. Eight-week engagements, OT-respecting integration, edge or cloud deployment shape, and an eval cadence that does not break a shift. SERVICE · 01 / 06 Asset reliability Predictive maintenance engine Time-series + LLM-on-event scoring that flags the asset before it fails. Drift-detected, retrained on a schedule, alerts the right OT engineer. - Sensor-fusion features with vendor-neutral schema - Drift detection per asset class + per supplier lot - On-call routing that respects shift schedules Python ClickHouse PostgreSQL OpenTelemetry SERVICE · 02 / 06 Line-speed inference Vision defect detection Real-time defect classification on the line. Edge inference where bandwidth or latency requires, cloud where flexibility wins. - Edge models tuned for line speed (sub-100ms per part) - Per-SKU eval set + drift alerts on new product runs - Operator-in-the-loop labeling pipeline for new defect classes Python ONNX Runtime Cloudflare Postgres SERVICE · 03 / 06 ISA-95-aware OT / IT bridge The middleware that lets your data science team see PLCs without giving them PLC access. Read-only by default, OT-engineer approved. - OPC-UA + MQTT ingestion with schema validation - Read-only abstraction layer between OT and IT - Per-line access controls that OT engineers can audit Python PostgreSQL OpenTelemetry gRPC SERVICE · 04 / 06 Run where it matters Edge inference + model registry Models on the line, on the gateway, or in the cloud. One registry, vendor-neutral, with rollback that does not require an OT engineer to drive. - ONNX runtime + Triton for production-grade edge throughput - Per-line model rollout with canary + auto-rollback - Hardware-aware compilation per gateway class ONNX Runtime Triton Kubernetes Cloudflare SERVICE · 05 / 06 What the plant manager sees Production observability + cost Per-line latency, drift alerts, energy-per-inference metering. Plant manager and CFO see the same numbers as the on-call. - OpenTelemetry across edge + cloud inference paths - Energy-per-inference budget where edge power is constrained - Eval-as-monitor: drift alerts before the line operator notices OpenTelemetry Grafana ClickHouse Sentry SERVICE · 06 / 06 Pre-build wedge Architecture review One-week audit of your manufacturing-AI stack. Edge hardware, model registry, OT bridge, eval cadence, drift response — all named in writing. - Edge hardware vs. cloud trade-off scored per line - Model registry shape: one-registry-rules-all or per-asset - Drift response cadence: who pages whom, on what threshold ADRs Triton ONNX Runtime OpenTelemetry Most engagements bundle two: an edge build (01, 02) paired with the discipline that keeps it shipping (03, 05). Bring the shape closest to your blocker. [Scope your engagement →](https://www.kensink.com/contact) Want to see the K-Framework discipline behind every item? [Read the K-Framework](https://www.kensink.com/k-framework). \[THE STACK · BY LAYER\] ## Boring infrastructure. Line-speed results. Tools your OT engineers have already approved. Self-host where edge requires, cloud where flexibility wins. LAYER · DATA + RETRIEVAL ### Data + retrieval. The store, the index, the search PostgreSQL pgvector Redis ClickHouse BigQuery OpenSearch LAYER · MODEL LAYER ### Model layer. Embeddings, providers, fallbacks OpenAI Anthropic Cohere Embed Voyage Llama (self-hosted) vLLM LAYER · EVAL + OBSERVABILITY ### Eval + observability. The eval bar, the cost meter, the drift alarm LangSmith Promptfoo OpenTelemetry Datadog Grafana Sentry LAYER · BACKEND + TRANSPORT ### Backend + transport. Type-safe everything TypeScript Next.js Python FastAPI gRPC BullMQ tRPC Zod LAYER · MOBILE ### Mobile. iOS + Android, native or cross React Native Expo Swift Kotlin FCM APNs LAYER · CLOUD + DEPLOYMENT ### Cloud + deployment. Whatever your infra already runs Cloudflare Workers Cloudflare R2 AWS GCP Vercel Fly ✕ WHAT WE DO NOT SHIP ### Direct against the model API. Edge inference where the line demands. - ✕ No LangChain - ✕ No LlamaIndex - ✕ No agent framework - ✕ No orchestration vendor - ✕ No black-box ML platform \[ PROOF · WHAT THE STACK DELIVERS \] ## Numbers that hold up at 24/7 line speed. MEASURED · WEIGHTED · 2024–2026 SCALE 10k + Sensors per deployment in production LATENCY p95 / 80 ms Edge inference round-trip target INTEGRATION ISA-95 OT integration default OPS 24/7 Production-grade uptime expectation MANUFACTURING AI · APPLIED K-FRAMEWORK ## Bring the floor problem. We'll bring the build. Eight weeks, fixed scope, eval suite + drift response at handoff. Direct LLM and ML engineering on top of the K-Framework. Two Q3 slots remain. [Start a project →](https://www.kensink.com/contact) [Read the K-Framework](https://www.kensink.com/k-framework) CYCLE 8 weeks · problem to line OUTPUT Code · evals · runbook DEPLOYMENT Edge or cloud