Baakey Dow

Identified an enterprise data gap for AI agents, designed the AIX Platform and drove its adoption as the Domain Gateways layer of Criteo's Sovereign AI initiative (Domains-Driven AI architecture). Sole technical lead on a 3-person core team. Authored 9 ADRs (architecture, security, vendor evaluations). MCP Platform • 4 servers, 76+ tools, 451 active users, 5,400+ invocations at 0% error rate during the measured window • Consumed by SAI/Agentfabric, Claude Code, Dust and Glean (~1,100 downstream users) • Shipped in 3 weeks (24 tools at launch, 76+ today), infra cost ~$3-5K/year (HTTP/KEDA autoscaling) Microsoft MCP server (47 tools) • Full Excel automation (workbook creation, formulas, charts, tables) • PowerPoint generation with hexagonal architecture (8 Criteo brand templates, OOXML manipulation) • SharePoint search and OneDrive file lifecycle (create, rename, delete, PDF conversion) Dust MCP server => meta-agent orchestration • Enables AI agents to programmatically create, configure and monitor other agents • Dynamic tool attachment, feedback-driven optimization, recursive delegation across Dust, Glean and SAI/Agentfabric @criteo/mcp-core • Shared library delivering ~80% code reuse => new server = 13-line bootstrap + 40-60 lines per tool • JWT validation, OBO client, OAuth proxy (RFC 8693), OpenTelemetry and JSON Schema tool contracts Identity and security • End-to-end identity propagation across 3 agent modes: autonomous (client credentials), on-behalf-of-user (OBO for Microsoft Graph, per-user OAuth for Zoom) and embedded (app identity) • Zoom build-vs-buy evaluation: deprecated custom server in favor of Zoom's 3 official MCP servers (18 tools), eliminating an entire deployment • GDPR/AI Act/French Labour Code (CSE consultation, art. L2312-38) analysis for Zoom transcription auto-logging • TypeScript vs MuleSoft technical evaluation for ARB (~380 lines TS vs 975 lines XML); integrated MuleSoft Flex Gateway as API governance layer Infrastructure and observability • Deployed to Azure Container Apps (DEV + PROD) via Terraform, 35 Azure resources across 2 subscriptions • 24-panel Azure Monitor workbook (per-user invocations, P50/P95/P99 latency, OBO token exchange, error forensics) auto-provisioned via Terraform • 10+ semantic-versioned releases (v0.1.0 → v0.10.2) in 4 months, cross-team coordination (Platform, EUS, Security, Architecture) Stack: TypeScript, Hono, MCP Protocol, Azure (Container Apps, Entra ID, Key Vault, ACR), Terraform, MuleSoft Flex Gateway, OpenTelemetry, Docker, Turborepo

Top contributeur de l'org GenAI Platform (300+ PR sur 15 repositories, 91 % de merge rate). 4 plateformes GenAI enterprise + 1 système de workflows IA développeur livrés en environnement pharma régulé. Enterprise Multi-Agent AI Gateway • Orchestration multi-agents en factory pattern (Bedrock, A2A Protocol avec parsing des agent cards et caching OAuth, REST, Flowise), SSE streaming 15 min (7,5x le timeout standard) • WAF v2 (SQLi, XSS, rate limiting), authentification 6 couches (Azure AD → JWT → OBO → API key → RBAC → ownership), pipeline PII bidirectionnel • Audit trails immuables DynamoDB (TTL 90 j, 3 GSI pour investigation forensique), publication SIEM via EventBridge • Observabilité 99,9 % sur 127+ Lambda (OpenTelemetry + Weave/W&B + Datadog) Pipeline de données non structurées (Python) • Step Functions 5 étapes déterministes (Partition → Extract → Chunk → Metadata → Vectorize), inférence multi-modèle Bedrock (Claude, Nova, Llama, Mistral, DeepSeek) • Pinecone + KMS à contexte cryptographique pour l'isolation tenant, évaluation LLM offline (Weave/W&B) • Remédiation proactive de 4 CVE critiques (pdfminer, torch, langchain-core, marshmallow) Context Manager Platform (10 000+ docs/mois) • Principal contributeur : dashboard CMS React 19 (TanStack Query, drag-drop, timeline temps réel, design system Sanofi) • Chiffrement KMS avec séparation des responsabilités (CMS chiffre, pipeline déchiffre via rôles IAM distincts), RBAC 3 niveaux Azure EntraID, validation Zod exhaustive • BYODS (Bring Your Own Data Source) avec ingestion event-driven S3, onboarding client −92 % (2 semaines → 4 heures) • Migration CDKTF en 6+ stacks modulaires (Base, DB, IAM, UDP, API, Event), déploiement indépendant avec state remote DynamoDB GenAIR + DocFactory + Platform Engineering • Pipeline event-driven (SNS/SQS/Step Functions, DLQ retry, zéro perte), transformation monorepo Turborepo (443 fichiers, 27K+ lignes) • Upgrade Redis (correctif CVE CVSS 10.0, +72 % de throughput, −71 % en P99), provisioning SageMaker Studio pour 50+ développeurs • 500+ tests automatisés (85 % de couverture), CI/CD smart (détection path-based, semantic versioning, promotion tag-based GitOps) Workflows IA développeur + Plugin Marketplace • 7 agents spécialisés avec quality gate obligatoire (frontend, backend-auditor, API-auditor, devops, quality-tester, UI-designer, Atlassian) • 6 serveurs MCP intégrés (GitHub, Atlassian, Confluence, Terraform, Figma, Playwright) avec auth token-based • Agents sécurité : détection de secrets, SQLi, failles de validation d'input ; agent IaC (Checkov, KMS, IAM least-privilege) • claude-box (CLI conteneurisée AWS Bedrock) + marketplace de plugins avec context engineering (~30k tokens économisés/interaction) Stack: TypeScript, Python, React 19, NestJS, Hono, GraphQL, FastAPI, AWS (Lambda, Step Functions, SageMaker, Bedrock, DynamoDB, SNS, SQS, S3, KMS, WAF, EventBridge), Terraform/CDKTF, Pinecone, OpenTelemetry, Docker

• Maintained mission-critical PIM platform serving 500+ product teams with 99.9% uptime during peak traffic • Led modernization initiative migrating Node.js 12->16, Yarn 1->3, Lerna 2->5, reducing build times by 25% and eliminating 47 security vulnerabilities • Delivered comprehensive technical documentation enabling 3x faster onboarding for new team members.

• Built ROI optimization platform for mobile ad campaigns (AppLovin, Vungle, IronSource, Unity Ads), enabling data-driven ROAS decisions across $10M+ annual ad spend • Developed 10+ API integration SDKs (TikTok, Facebook, Snapchat, AdMob, Mintegral) for authentication, reporting, and advertisement operations • Led migration to Lerna monorepo architecture and Redux -> Context API refactoring, improving developer productivity by 30% • Orchestrated backend data migration from Redshift to BigQuery, reducing query costs by 40%

• Bootstrapped enterprise BI application (Google Data Studio alternative) with CAS/LDAP Spring Security authentication • Designed dynamic plugin infrastructure using webpack for runtime JavaScript bundle loading • Led CI/CD implementation (Docker, GitLab runners, Maven), reducing deployment time by 60%

• OAuth integrations for 4 major platforms (Dropbox, Twitter, Google Drive, Instagram) • WYSIWYG text editor in React with rich formatting capabilities • Dropbox API v1 -> v2 migration ensuring zero downtime. • Add new features, enhance existing functionalities, and initiate new projects. • Identify integration points with existing internal and external systems to facilitate seamless data exchange and interoperability between systems. • Ensure applications are robust, well-structured, and adhere to industry best practices and coding standards. • Implement measures to guarantee the reliability and scalability of the developed applications, allowing them to handle increased user loads and adapt to future growth requirements.