The quiet link between data platforms, the teams that rely on them, and the problems that try to take them down.
Yahya Bloukh — Data Platform Architect & AI Solutions Engineer
Keep the data flowing, keep the teams unblocked, and keep the things that try to take it all down from winning. The AI work grows out of the same instinct — build the system, then teach it to run itself.
Design, deploy, and operate enterprise data platforms end to end — Cloudera, Greenplum, the Qlik Talend suite, and Tableau. Uptime ownership, vendor-led upgrade planning and execution, and cluster sizing and PoCs across on-prem and private cloud. Security-aware by background: vulnerability remediation, RBAC and Apache Ranger access management, and deployments built to satisfy InfoSec, network, and compliance.
The last line of defense when production breaks. Diagnosing complex issues across the stack, unblocking data engineering teams, and supporting data science with integrations and non-standard solutions. The strength: breaking big problems into small ones, fast.
Agentic AI systems built with CrewAI, LangChain/LangGraph, and RAG architectures. Ongoing research into on-prem LLM deployment — GPU selection per model class, and realistic throughput benchmarking for production workloads.
3× enterprise GenAI hackathon winnerAn on-prem extension to observability platforms that automates the safe fixes and tracks every incident end to end — giving first-line operators fewer escalations, faster recovery, and an RCA that's already written.
View case study →Three first-place finishes in enterprise hackathons at a national telecom — all agentic-AI use cases: multi-agent systems with CrewAI, orchestrated workflows with LangChain/LangGraph, and RAG pipelines.
A practical mapping of GPUs to model classes: what hardware hosts which models, and what throughput each combination realistically sustains for standard tasks — the numbers you actually need before deploying on-prem.
Cluster sizing and proof-of-concept engagements with Cloudera, for both on-prem and private-cloud deployments — turning capacity questions into defensible architecture decisions.
A web app for running data-quality checks against different databases — turning ad-hoc validation into something repeatable.
Lets business users manage their own tables and lookups directly — no engineering ticket, no waiting in the queue.
A configurable, regex-driven file-routing engine moving files between systems: local↔HDFS and server-to-server.
Solo architect operating at team-lead grade, responsible for the company's core data platforms — administering Cloudera Data Platform, Greenplum, Talend, and Tableau environments. Leads agentic-AI use cases including on-prem LLM deployments, MLOps pipelines, and model evaluation; executes platform PoCs and multi-node installations; contributes to the legacy-EDW migration onto Cloudera and Greenplum; builds internal governance, lineage, and compliance tooling; remediates InfoSec-reported vulnerabilities and enforces RBAC; and acts as the final escalation point for data engineering teams.
Built production tooling in Python and Bash, with daily work across Spark, Airflow, Presto, Hive, and HBase on RedHat/CentOS. Designed feed-ingestion models and ETLs across multiple sources, resolved internal and client escalations, and built reporting dashboards in Power BI.
Vulnerability assessments and penetration testing, configured and deployed AlienVault OSSIM SIEM, and delivered client proofs-of-concept. The security instinct still shapes how every platform gets built.
Available for sizing, on-prem deployment for data-sovereignty requirements, third-line escalation, and consulting — or a full-time role for the right team. Let's talk.
An on-premise extension to observability platforms, built to back up first-line (L1) operators: it automates the remediations that are safe to automate and records the full lifecycle of every incident — so each one becomes a root-cause analysis that's written by the time you read it.
Failures that follow a known runbook are remediated the moment the platform flags them — resolved before they reach an operator's queue, so the team stays focused on incidents that actually need judgment.
Automation is scoped to safe, reversible actions. Anything destructive — or outside a known-good runbook — stops and waits for an operator to approve it, so the system never makes a bad situation worse.
Every incident is tracked end to end: when it started, what closed it, and each step along the way — the actions the system ran automatically, or the steps the operator took when a fix needed a human.
Because the timeline and actions are captured as the incident unfolds, the root-cause analysis assembles itself — no reconstructing what happened after the fact.