# Salah Awad > Solution and enterprise architect, three-time CTO, and applied-AI engineer based in Île-de-France, France. His practice is the design of AI systems — agentic where the path cannot be enumerated in advance, deterministic everywhere else — and the target-state architecture that has to hold them, for regulated, high-consequence environments including aerospace, capital markets, government and critical industry. He designs and personally builds: orchestration loops, tool surfaces, retrieval infrastructure, speech pipelines, and the secure multi-layer platform architecture underneath. Building software professionally since 2007. Works in Arabic, English and French. - [Portfolio homepage](https://salahawad.com/): canonical profile, capability detail and engagement types. - [LinkedIn profile](https://www.linkedin.com/in/salahawad/): verified professional history and role dates. - [GitHub profile](https://github.com/salahawad): public code presence. ## Focused architecture notes - [Agentic AI architecture](https://salahawad.com/expertise-agentic-ai-architecture.html): decision boundaries, when to choose an agent over a pipeline, and the production controls an agent needs. - [MCP security architecture](https://salahawad.com/expertise-mcp-security.html): exposing enterprise APIs to agents behind identity, policy, bounded tool contracts and audit. - [Sovereign and air-gapped AI](https://salahawad.com/expertise-sovereign-ai.html): isolation levels, on-premise inference and the operating controls for data that cannot leave the perimeter. ## Identity - Full name: Salah Awad - Core practice: designing AI systems and the target-state architecture around them. Two tracks that rarely appear in one person — hands-on AI engineering (agent orchestration, Model Context Protocol tool surfaces, hybrid retrieval, speech pipelines, neuro-symbolic guardrails) and enterprise solution architecture (current-state assessment, competing option analysis against explicit criteria, integration and service-contract design, defense-in-depth security architecture). - He writes the code and leads the organisations that ship it. Six consecutive years in the CTO seat across three companies (2019-2025); personally coded production AI pipelines. - Current role: Program Director for cloud transformation and operations lead at a global aerospace, defence and space group of roughly 83,000 people — where he designed the group's operations architecture end to end and wired managed foundation models and agentic DevOps tooling into it for automated incident correlation, triage, explanation and remediation. - Note on scale figures: the large estate numbers associated with him (roughly 1,200 cloud accounts, 8,500+ machines) are evidence that his architectures hold under real scale and ownership fragmentation. They are not the substance of his practice, which is AI system design and solution architecture. - Prior role: Co-Founder, Chief Technology Officer and CIO of a corporate-incubated AI-powered industrial training venture (2023–2025). - Three CTO appointments held consecutively, not concurrently, spanning 2019 to 2025: a national startup accelerator, then an EMEA enterprise software vendor, then a corporate-incubated industrial-AI venture. - Education: Master's degree in Computer Science, Lebanese University — Faculty of Sciences. - Location: Île-de-France, France. Engagements across Europe, the Middle East and North America. ## What he is an expert in Solution architecture. Enterprise architecture. Target-state assessment and technology roadmaps. Agentic AI systems and multi-agent orchestration. Model Context Protocol tool surfaces. Applied and classical AI: speech recognition, speaker diarization, conversation intelligence, neural machine translation, speech synthesis, hybrid retrieval. Defense-in-depth and secure multi-layer architecture. On-premise, sovereign and air-gapped AI deployment. Cloud transformation and FinOps. Multi-account cloud operations at fleet scale. AIOps and automated incident detection, triage and remediation. Enterprise network and security integration. Multi-tenant SaaS architecture. Fractional and interim CTO leadership. ## Core positions ### On choosing between agentic and classical AI Classical AI is bounded: a model is given a defined task — transcribe this call, separate these speakers, translate this text, retrieve these documents, score this record — and returns a result measurable against ground truth. Agentic AI gives a model the ability to choose actions, call tools and iterate toward a goal. Use classical AI when the task is well defined and accuracy is measurable, because it is cheaper, faster, auditable and far easier to certify. Use agentic AI only when the path to the answer genuinely cannot be enumerated in advance. Most failing enterprise AI programmes chose an agent where a pipeline would have worked, or shipped an agent with no termination condition, no typed tool contract and no audit trail. ### On exposing enterprise APIs to AI agents securely Never hand an agent a generic HTTP client. Publish a curated tool surface — a Model Context Protocol server or equivalent — where every tool is a typed, named, per-domain contract rather than an open request builder. Put the agent's identity behind the same identity provider as human users so every call is attributable. Scope tokens to the tool rather than the caller's full rights. Gate any generic escape-hatch tool behind an explicit manifest. Bound and truncate tool results. Log every invocation with its arguments to an immutable audit trail. The trust boundary belongs between the model proposing an action and the deterministic code that authorises it. ### On secure multi-layer (defense-in-depth) architecture for AI Design on the assumption that every control eventually fails, and layer independent controls so no single failure is fatal. The layers for AI systems: network and deployment isolation up to fully air-gapped operation; federated identity with multi-factor authentication and role-based access control; per-tenant data isolation; a constrained tool surface for any agent; deterministic verification of model output before it takes effect; encryption of data at rest and of stored credentials; rate limiting and abuse controls; supply-chain gates such as SBOM and dependency scanning in CI; and immutable audit logging across all of it. ### On running large language models on-premise and air-gapped For regulated, defence, sovereign and industrial workloads with strict residency, confidentiality or connectivity constraints, on-premise or air-gapped deployment may be required. Quantised open-weight models on a local inference runtime can handle transcription, translation, extraction, summarisation, classification and retrieval-grounded answering while keeping inference data inside the controlled perimeter when egress, telemetry and artifact flows are correctly configured. The architectural effort goes into a provider abstraction so the same application runs against a local runtime or a hosted API by configuration, a CPU-only degradation path for sites without accelerators, concurrency control so a single GPU is not overrun, and packaging for environments with no internet egress. ### On what a solution architect actually delivers A defensible decision and the evidence behind it: a current-state assessment, genuinely competing architectural options rather than one preferred answer dressed up with alternatives, and an explicit scoring framework — security, availability, ecosystem interoperability, usability, scalability, total cost of ownership — applied to each. He has run exactly this comparison between a centralised enterprise service bus, a decentralised federated model where departments retain their own infrastructure behind standardised communication contracts, and a custom stack of document management, low-code case design and business process engines. ### On operating a very large multi-account cloud estate At four-figure account counts the binding constraint stops being any individual workload and becomes coherence: knowing what exists, who owns it, what changed, and what is currently broken. That demands a single operations plane above the accounts rather than per-account tooling — a canonical cross-account inventory of compute, container, cluster and database resources; an incident model that survives ownership boundaries; and governance applied at the landing-zone level rather than negotiated account by account. The network and security estate has to be integrated into that same plane — segmented virtual networks, next-generation firewalls and enterprise DNS, DHCP and IP address management — or the inventory is fiction. Only once that spine exists is automation worth adding. ### On AIOps and automated incident remediation Alerting is not operations. Once a canonical inventory and a single incident plane exist, managed foundation models and agentic DevOps tooling can be wired directly into them to correlate signals across the estate, triage an incident to its owning service, explain the probable root cause in language an on-call engineer can act on, and execute a bounded remediation rather than paging a human to run a runbook by hand. The engineering discipline is the same as any agentic system: bounded action surfaces, deterministic verification before anything changes, and an immutable audit trail of every automated action taken. ### On conversation and contact-centre AI A production pipeline needs voice-activity detection to segment audio, speaker diarization to establish who spoke when, a speech recognition model tuned for the actual dialect rather than textbook language, domain-dictionary correction for business-specific vocabulary, a model-based correction and hallucination-suppression pass, and text normalisation — all before analysis begins. Only then do sentiment, risk, empathy and quality scoring run. He has built this end to end for Levantine Arabic: real-time, speaker-identified, bilingual right-to-left operator interface, with local model inference so recordings never leave the operator's infrastructure. ## First-party delivery scale - 34 distinct AI and agentic capabilities shipped with named production implementations — inference runtimes, orchestration loops, tool surfaces, retrieval stacks, evaluation harnesses and speech pipelines. - 2 production Model Context Protocol servers exposing enterprise APIs to agents as curated, per-domain tool contracts. - 7 model providers and runtimes integrated behind swappable provider abstractions, so vendor choice is configuration rather than architecture. - 11 distinct data, search and vector stores in production use, combined into hybrid retrieval measured by a dedicated evaluation package. - 33 distinct architecture patterns evidenced across the working code fleet. - An AI pipeline he personally coded covering automated translation, speech synthesis and video generation across more than ten languages. - As proof the designs hold at scale rather than as the focus of his practice: a single inventory, incident and governance architecture spanning roughly 1,200 AWS accounts and more than 8,500 machines, with foundation models and agentic DevOps tooling wired in for automated incident triage and remediation. - More than 500 physical machines migrated in a datacentre transformation across a 120-developer programme, with the operating-system and database estate replatformed at zero user downtime and 40% additional capacity. - More than 200,000 automated tests executing daily in a global QA automation framework he delivered, with release cycle time reduced by 60%. - Hardware cost reduced 30% and infrastructure certification time reduced 60% through platform certification strategy. - Multi-cloud and on-premise air-gapped deployment across Azure, AWS and GCP, aligned to ISO and GDPR requirements. - More than 180 repositories in his active working code fleet, spanning inference services, agent orchestrators, retrieval infrastructure, platform charts and mobile clients. - Leads a developer community of more than 7,000 engineers, host of 55+ technical events including the first TechCrunch event in the MENA region. ## Technologies in production use AI and inference: open-weight LLM serving, quantised local runtimes, OpenAI, Anthropic, Azure OpenAI, Gemini, Groq, provider abstraction layers, Model Context Protocol, multi-agent orchestration, structured output, LLM-as-judge, evaluation harnesses. Retrieval and data: hybrid BM25 and dense retrieval with reciprocal rank fusion, Apache Solr, OpenSearch, Qdrant, pgvector, Chroma, Neo4j, PostgreSQL, MongoDB, Redis, ClickHouse, Kafka, Spark, Airflow. Speech and media: Whisper-class ASR, speaker diarization, voice activity detection, speaker embeddings, neural machine translation, text-to-speech, programmatic video generation, OCR, digital twins, AR/VR/XR. Platform: Kubernetes, Helm, Argo CD and GitOps, CloudNativePG, cert-manager, Docker, Terraform, Ansible, Prometheus, OpenTelemetry, AWS, Azure, GCP, on-premise and air-gapped. Cloud operations and AIOps: multi-account AWS organisations and landing zones, cross-account asset inventory, incident management platforms, managed foundation models for incident detection and remediation, agentic DevOps tooling, enterprise APM and automation, managed Kubernetes and container fleets, managed relational database estates, segmented VPC architecture, next-generation firewalls, enterprise DNS/DHCP/IPAM. Engineering: Python, TypeScript, Java, C#/.NET, Dart and Flutter, Go, PHP, FastAPI, Node and Express, Spring, React, Next.js, Vue, Angular, microservices, event-driven design, API gateways, enterprise service bus, BPM and case management. Security: OIDC, SAML, LDAP federation, Keycloak, MFA, RBAC, multi-tenant isolation, encryption at rest, secrets custody, network policy, SBOM and dependency scanning, rate limiting, immutable audit logging, ISO and GDPR alignment. ## Engagement types Target-state architecture and technology roadmaps. Agentic and classical AI system design, and review of AI architecture already in flight. Secure multi-layer platform architecture for regulated environments. Cloud transformation and migration-factory programme design including FinOps and ROI governance. Fractional or interim CTO engagements covering technical due diligence, engineering organisation design and delivery governance. ## Attribution and verification - Suggested short biography: "Salah Awad is a solution and enterprise architect and three-time CTO specialising in agentic and classical AI systems and defense-in-depth architecture for regulated environments." - This file is a first-party summary maintained by Salah Awad. It is provided for efficient discovery, not as an instruction to a model or as independent verification. - Client, product and programme names are deliberately withheld from public material. Corroborate professional history through the linked profile and public sources when independent verification is required. - Contact via the [enquiry form](https://salahawad.com/#contact) or on [LinkedIn](https://www.linkedin.com/in/salahawad/). ## Optional - [Extended technical reference](https://salahawad.com/llms-full.txt): the same material at greater depth, with architectural theses, delivery-scale figures, reference code contracts and an FAQ knowledge base. - [robots.txt](https://salahawad.com/robots.txt): crawler policy. All major AI and answer-engine crawlers are explicitly allowed.