Solutions Architect

Solution Architect, Enterprise Video Analytics Platform Function: Technology / Platform Engineering Experience: 12+ years Location: Abu Dhabi, UAE Travel: Frequent international travel expected, primarily to India About the Role We are building and operating one of the largest AI video analytics platforms in the region. The platform runs computer vision workloads across a distributed camera estate spanning thousands of operational sites, deployed on a hybrid footprint where most inference happens at the edge and the control plane, data platform and centralised workloads run on Azure. Multiple business domains run on it concurrently, from consumer retail and specialty retail through to telecommunications infrastructure and heavy process industry. At this size, the hard problems are not the models. They are throughput per accelerator, fleet management across a large population of remote devices, tenant isolation, event volumes, storage growth, integration into systems we do not control, and knowing whether tens of thousands of endpoints are actually healthy without a human looking. As Solution Architect you own the end to end technical design of that platform: edge inference, video ingestion, model serving, event distribution, data platform and enterprise integration. You set the architecture, you own the numbers behind it, and you are accountable for it continuing to hold as the estate grows. You work closely with delivery squads, platform engineering, data science and enterprise stakeholders, and you act as the senior technical voice in architecture governance. Responsibilities Architecture at scale Own the target state architecture for the platform and the standard deployment patterns applied across business domains, including where they must legitimately diverge and where divergence has to be prevented. Own the multi tenancy architecture: isolation across data, models, identity, compute and cost attribution, including physically separated tenancy where required. Multiple domains depend on this, so it sits on the critical path. Drive a reusable use case and dashboard framework so new domains and analytics are configured rather than rebuilt from scratch each time. Define the hybrid edge and cloud placement model: what infers at the site, what streams centrally, what is retained where, and how each site degrades independently when connectivity or hardware fails. Define the workload tiering strategy across continuous inference, event triggered, batch and archive only. At this estate size this single decision drives most of the cost and most of the feasibility. Set the non functional baseline: availability targets, latency budgets per use case, RPO and RTO, failover behaviour, and the failure modes the design explicitly does not cover. Own architecture decision records, high and low level design documentation, and the architectural risk and assumption log. Edge and GPU infrastructure Own the edge reference architecture: device and accelerator selection, GPU sizing, hardware accelerated decode, thermal and power constraints, site networking, and remote recoverability. Own edge fleet orchestration and device management at scale: provisioning, configuration, health telemetry, remote diagnostics, and safe progressive rollout of software and models across a large distributed device population. Lead model porting and optimisation across heterogeneous silicon, spanning data centre GPU generations and embedded edge platforms, and own the benchmarking methodology and migration decisions. Own inference throughput economics: streams per accelerator, batching, multi model scheduling on shared GPUs, quantisation and runtime optimisation, with measurable targets and evidence behind them. Own GPU and edge capacity planning against the rollout plan, including procurement lead times. Computer vision and model lifecycle Translate business requirements into computer vision solution designs across safety, security, operational compliance and customer analytics. Typical workloads include PPE and coverall compliance, personnel counting across multiple uniform types, restricted zone and access violation, vehicle and forklift safety, fire and smoke detection, slip and fall, intrusion, camera tampering, procedural (SOP) compliance, queue and dwell analytics, footfall and zone flow mapping, and housekeeping verification. Own the model lifecycle architecture: dataset and annotation pipelines, training and evaluation infrastructure, model registry, staged promotion, canary rollout across sites, drift detection and retraining triggers. Define accuracy and false positive acceptance criteria per use case, along with the measurement method. Operator trust at scale is lost to alert fatigue long before it is lost to missed detections, so design and argue accordingly. Own responsible AI review for sensitive workloads. Demographic estimation, emotion inference and personnel monitoring carry bias, validity and regulatory exposure, and you are expected to surface those constraints early rather than after deployment. Integration architecture Own the integration architecture and interface catalogue across: VMS platforms: enterprise video management system connectivity at estate scale, ONVIF, RTSP and equivalent streaming protocols, and mixed vendor camera estates. Enterprise systems: ERP and asset management, ITSM and incident routing, identity (Entra ID, LDAP), notification and mobile field applications. Data and analytics: lakehouse and pipeline integration, BI, and product usage instrumentation. Domain specific systems: OT and ICS environments in industrial settings, POS and loss prevention in retail, network operations and site access systems in telecommunications. Define canonical event and metadata schemas so downstream consumers are decoupled from model internals and from VMS vendor formats. Set integration patterns and governance: event driven and asynchronous design, idempotency, backpressure, replay, versioning and deprecation. Own automated site and camera onboarding, including automated zone configuration through client APIs. Manual onboarding does not scale, and the automation path is yours to design and drive. Data platform Own the data architecture: ingestion, lakehouse and pipeline design, retention tiering, archival and compression, and the growth model behind all of it. Own data governance and privacy architecture for video, biometric and derived personal data, including applicable data protection obligations. Reliability, observability and operations Own the observability architecture end to end: metrics, logs, traces, and critically, camera level, site level and device level health telemetry. At this scale, "is it working" must be answerable without human inspection. Define SLIs and SLOs, alerting philosophy, escalation paths, and the on call and runbook model handed to the operations organisation. Own the deployment and release architecture: infrastructure as code, GitOps, progressive delivery, environment topology, and how a change reaches every site safely without causing a fleet wide incident. Own FinOps and unit economics: cost per camera per use case, cost guardrails, and the levers available when the platform bill needs to come down. Stay engaged through operational handover and steady state rather than stopping at go live. Technical leadership and stakeholder engagement Act as the senior technical authority in architecture governance and design authority forums, internally and with enterprise stakeholders. Provide technical leadership, mentoring and design review across delivery squads, partner teams and client engineering teams. Own technical dependency management. A meaningful share of delivery risk sits with external parties, covering sample data, stream access, integration endpoints, platform decisions and site access. You identify, document, chase and escalate these with clear impact statements rather than absorbing the slippage. Support presales an