Full-Stack + AI/ML + Data Science + Agentic
IC · Technology Leader · Engineering Manager — whichever seat the problem needs. Not mutually exclusive, usually juggling .
Principal Software Engineer on Red Hat's Data & AI team.
I build and run AI Factory — the internal agentic platform Red Hat engineers use to get agents into production. Control plane, edge gateway, and the agent + UI templates every agent on it starts from. I designed its agent evaluation framework. My team builds it; I'm in the code with them.
I came to agents through the models, not around them. Six years at Ericsson R&D — data scientist first: exploratory analysis, anomaly detection tuned for telco data, machine reasoning, model selection, monitoring, retraining, the whole model lifecycle. Then the layer underneath: architected a cloud-native MLOps platform from scratch and owned it as product owner, then led the engine that took it to parity with Vertex AI and SageMaker. MLflow, Kubeflow, Seldon, KServe, Argo — the plumbing that carries a model from ideation to industrialization. Traditional AI and GenAI both; I didn't skip the first to get to the second.
Now the same discipline one layer up. RAG and Graph RAG, vector stores, LangGraph and MCP, fine-tuning, SLM-vs-LLM benchmarking, LLM-as-judge evaluation. Agentic systems are still ML systems — same rigour needed on data, evaluation and drift — with one new problem: they're probabilistic apps, and they fail in ways deterministic software never did.
Which is why I spend as much time on reliability as on features: chaos testing, circuit breakers, and the unglamorous work that keeps agentic systems standing up.
Seventeen years to get here, and I've been the IC, the lead and the manager along the way — often at the same time. The through-line is production, not demos.
Why this matters now. Agents have democratized code generation and UI work — the low-hanging GenAI use cases are already picked. What's left needs enterprise architecture thinking, proven design patterns and production-grade engineering discipline. That's the part I spent seventeen years learning, one layer at a time.
Top of the stack down to the foundations — the order I learned it in reverse.
agentic:
frameworks: LangGraph | LangChain | Deep Agents | Strands | MCP (server + client)
platform: control plane | edge gateway | agent & UI templates | Aegra runtime
retrieval: RAG | Graph RAG | vector stores | ONNX embeddings
evaluation: LLM-as-judge | Promptfoo | Lightspeed benchmarks | Custom Agentic eval harness
models: Claude | Gemini | open-source SLMs | fine-tuning | SLM-vs-LLM benchmarking
governance: OPA policy | guardrails | HITL approval | AI security
dev_tooling: Claude Code | Kiro CLI | agent skill authoring | enterprise skill builder
ai_ml:
data_science: EDA | anomaly detection | correlation | machine reasoning | feature analysis
modelling: model selection & tuning | monitoring | retraining | drift | full model LCM
mlops: MLflow | Kubeflow | Seldon | KServe | Argo Workflows
clouds: AWS SageMaker | Bedrock AgentCore | GCP Vertex AI | Google Agent Engine
data:
elk: Elasticsearch | Logstash | Beats | OpenSearch
pipelines: Kafka | RabbitMQ | PySpark
stores: PostgreSQL | MongoDB | Valkey | Redis
warehouse: Snowflake | S3 | Oracle | CouchDB | SQLite
platform:
orchestration: Kubernetes | OpenShift | Helm | Kustomize | Argo CD | Argo Rollouts | Tekton
scaling: HPA | KEDA | blue-green | multi-cluster routing
local: Kind | Podman | Docker
reliability:
chaos: Toxiproxy | fault injection | blue-deployment isolation
patterns: circuit breakers | distributed semaphores | leader election
graceful drain | reconciliation loops | connection pooling
observability: OpenTelemetry | Prometheus | Grafana | Langfuse
Sumo Logic | Datadog | structlog | Alertmanager
security:
identity: SSO | OIDC | OAuth2 + DCR | Keycloak | LDAP | RBAC / ABAC
secrets: HashiCorp Vault | ArgoCD Vault Plugin
appsec: SSRF protection | CVE remediation | four-eyes approval | audit logging
engineering:
languages: Python | Java | JavaScript | SQL | Bash
backend: FastAPI | Pydantic | Spring Boot | Spring MVC | Node | Express
frontend: React | PatternFly | Redux | Angular | Fastify BFF | SSE streaming
testing: pytest | Playwright | Selenium | BDD | chaos suites
ci_cd: GitHub Actions | GitLab CI | Tekton | Argo CDBoth traversals. Breadth is the number of rows — depth is what's inside them.
| Layer | Where I've gone deep |
|---|---|
| Agentic platforms | Designed the agent evaluation framework. Own the control plane, edge gateway and the agent + UI templates every agent on the platform is built from. |
| AI / ML | Anomaly detection tuned for telco data, correlation algorithms, machine reasoning, model selection and tuning, full model lifecycle. |
| ML platform · MLOps | Architected a cloud-native MLOps platform from scratch and owned it as product owner. Led the engine that took it to parity with Vertex AI and SageMaker. |
| Reliability | Built a declarative chaos / fault-injection framework for Kubernetes. Circuit breakers, concurrency and lifecycle safety, reconciliation, production hardening. |
| Full-stack | MEAN-stack enterprise applications, Angular admin dashboards, orchestration pipelines, BFF and SSE streaming. |
| Testing · automation | Automation and migration frameworks, SDET tooling, reverse-engineered a vendor's proprietary test format into a reusable parser. |
Seventeen years, one layer at a time — and I still write the code in all of them.
| Repository | What it is | Role | Stars |
|---|---|---|---|
| template-agent | LangGraph agent template with enterprise security and Kubernetes deployment built in | Maintainer |  |
| template-ui | Agent UI that works out of the box with template-agent, extensible to any AI agent | Maintainer |  |
flowchart BT
subgraph IBM["IBM · 2009 — 2020"]
direction LR
A1["Test<br/>Specialist"] --> A2["Test Automation<br/>Specialist"] --> A3["Automation Developer<br/>SDET"] --> A4["Full-Stack<br/>Developer"] --> A5["Application Architect<br/>+ Analytics"] --> A6["Technical<br/>Manager"]
end
subgraph ERI["Ericsson · 2020 — 2026"]
direction LR
B1["Data<br/>Scientist"] --> B2["AIML Architect<br/>MLOps Engineer"] --> B3["AIML Solutions Architect<br/>MLOps Platform Engineering"] --> B4["AI-First Technology Leader<br/>AI Architect Manager"]
end
subgraph RH["Red Hat · 2026 — now"]
direction LR
C1["Principal Software Engineer · Data and AI"]
end
IBM --> ERI --> RH
classDef now stroke:#EE0000,stroke-width:3px
classDef ai stroke:#00D9FF,stroke-width:2px
class C1 now
class B1,B2,B3,B4 ai
style IBM fill:#ebedf0,stroke:#9aa4ae,color:#24292f
style ERI fill:#ebedf0,stroke:#9aa4ae,color:#24292f
style RH fill:#ebedf0,stroke:#9aa4ae,color:#24292f
| Years | Role | |
|---|---|---|
 |
2026 — Present | Principal Software Engineer · Data & AI Agentic platform engineering — control plane, edge gateway, agent & UI templates, agent evaluation framework, and the reliability work that keeps it in production. |
 |
2020 — 2026 | AI-First Technology Leader · AI Architect Manager · 2025–26 AIML Solutions Architect · MLOps Platform Engineering · 2022–25 AIML Architect · MLOps Engineer · 2022 Data Scientist · 2020–21 Cloud-native MLOps platform built from scratch and owned as PO. Anomaly detection and correlation for telco data. GenAI and traditional AI use cases taken to production. |
 |
2009 — 2020 | Technical Manager · Full-Stack Architect Application Architect + Analytics Full-Stack Application Developer Automation Developer · SDET Test Specialist Eleven years, six rungs. MEAN-stack enterprise applications, automation and migration frameworks, Python/Pandas analytics tooling. Onsite in the US as client SPOC. |
Red Hat is an IBM entity — Red Hat through Sep 2026, IBM CIO from Oct 2026, same team and same platform throughout.
Nearly a decade in Data & AI. Currently building large-scale, enterprise agentic systems for production.
Happy to talk.