Senior .NET & Data Engineer  ·  Athens, Greece

I build production-grade .NET platforms — data pipelines, multi-tenant SaaS, ML systems, and native Windows apps — with the full HA infrastructure behind them.

📇 Portfolio index  ·  💼 LinkedIn  ·  ✉️ Email

I design and ship end-to-end .NET / C# systems on production-grade infrastructure — SQL Server Always On, Kafka, StarRocks, ClickHouse, Percona MySQL, Kubernetes, Docker Swarm, HashiCorp Nomad/Consul/Vault — with full observability and documented runbooks.

The work is organised as the NexusPlatform portfolio: 14 application projects, 10 built infrastructure repositories, and a 4-app native Windows suite, each built to senior-engineer standards. See the portfolio index for the complete project grid with live status.

Phase 0 — Infrastructure foundation (COMPLETE) The 140-VM lab is fully built + cold-rebuild-proven end-to-end. Foundation tier (AD DS + HashiCorp Vault HA on Raft + PKI + LDAPS + Transit auto-unseal), orchestration tier (Docker Swarm + HashiCorp Nomad/Consul + Portainer CE, mTLS end-to-end), the Kafka ecosystem tier (two KRaft clusters on mutual TLS + Schema Registry + Connect + Debezium + ksqlDB + MirrorMaker 2 cross-cluster DR), and the OLTP data tier (6-node Redis 8.0 Cluster + 3-node MongoDB 8.0 Replica Set + 3-node Percona XtraDB Cluster + 2-node ProxySQL with VRRP-floated VIP, all on mTLS via Vault PKI, all cold- rebuild-proven end-to-end via per-engine Packer templates + per-cluster Terraform states) are all live. The .NET 10 Native AOT operator CLI (nexus-cli) ships all 5 of 5 master-plan verbs as of v0.5.0 — Phase 0.F closed; live RTOs all under master-plan budgets. Now at v0.8.12 (2026-07-10) — nexus-cli 🏁 100% COMPLETE: all 13 GAPs closed + infra-hardening 4/4 done + GENUINE-NA rulings reviewed+confirmed (330 tests, AOT 28.34 MB). The infra-hardening block delivered the 0.N.1 sharded-MongoDB wire mTLS (requireTLS + per-host mongo-sharded-server Vault-PKI leaves + online rotateCertificates via MongoShardedAdapter cert-rotate) and the 0.O.1 Vitess engine-native file BackupStorage on NFS + xtrabackup (VitessAdapter rewritten), plus the platform CA rollover (all tiers re-rooted, complete 2026-07-04/05). Earlier v0.8.7 (2026-07-06) — the completion backlog closed the last in-CLI gaps: scale-up/VmrunVmResizer (vertical resize + a cluster-safety gate + the honest deb13 root-not-last disk warning), swarm guarded backup restore --confirm-destructive, the kafka controller-leader resize-gate, and FoundationAD backup take (ntdsutil ifm) + FSMO failover-test. v0.8.5 (2026-06-26) — Phase 0.L.4: the Registry adapter (ClusterId registry) — the fifth and LAST non-data-tier adapter, completing full IClusterAdapter coverage of the fleet — the CLI now deeply manages EVERY cluster (all 5 non-data tiers + all 13 data/sharded families). Manages the highly-available Harbor container registry: 2 stateless Harbor app nodes (round-robin DNS registry.nexus.lab) + a dedicated PostgreSQL/Redis HA datastore pair behind a VRRP VIP (registry-db.nexus.lab .119) + image blobs in MinIO (s3://harbor) + Vault as the OIDC SSO provider (login delegated to Vault → AD). Same SSH-local-curl posture as obs/lakehouse (each node's own ca; Harbor API over node-local HTTPS; PG/Redis/keepalived via SSH; KV via INexusVaultClient); no managed Harbor/Docker-registry/Npgsql driver (NetArchTest). Verbs live-verified GREEN + cold-rebuild-proven: status (2 app nodes + datastore pair + VIP holder) · health (both Harbor cores up via RR DNS, PG streaming replication, Redis reachable, MinIO s3://harbor reachable, Vault OIDC issuer live, VIP bound) · topology · failover = registry-db VRRP cutover (keepalived) · scale-out = stateless Harbor app node add/remove (fixed-HA datastore → graceful N/A) · cert-rotate (per-node registry-server Vault PKI + reload) · acl (Harbor projects/robot-accounts + Vault-OIDC group mapping) · backup (PG dump round-trip + the MinIO-resident blobs) · chaos (process-kill a Harbor app node — the RR-DNS pair tolerates 1). 28.04 MB AOT; 243/243 tests; ADR-0026. 5 of 5 non-data tiers done — full-fleet adapter coverage achieved (all 5 non-data tiers + all 13 data/sharded families). Preceded by v0.8.4 (2026-06-25) — Phase 0.L: the Lakehouse adapter (ClusterId lakehouse) — the fourth non-data-tier adapter and the last big multi-component one: ONE component-aware adapter spanning the three-engine lakehouse (MinIO erasure-coded object store + Iceberg/Nessie REST catalog + Spark ZK-HA) plus the ZooKeeper ensemble across 16 VMs + 1 VRRP VIP. Same SSH-local-curl posture as obs (each node's own ca; Nessie mgmt + Spark UI plain HTTP; MinIO admin via the on-node mc nexuslocal alias; KV via INexusVaultClient); no managed MinIO/Spark/Iceberg/Nessie driver (NetArchTest). 11 verbs live-verified GREEN as-is + the full matrix green post-rebuild: status (16 nodes + VIP holder + ALIVE Spark master) · health (MinIO live/cluster/drives, Nessie object-store + catalog, Spark ALIVE + workers, ZK quorum, iceberg-pg replication, VIP) · topology · failover --direction spark-master (ZooKeeper auto-promotes the STANDBY master, RTO≈31s; --direction iceberg-pg = graceful N/A — a VRRP cutover split-brains the catalog DB) · scale-out N/A · cert-rotate (MinIO big-bang + Nessie; Spark/ZK N/A — no rotatable leaf) · acl (MinIO policies + users via mc admin) · backup (mc mirror s3://warehouse round-trip) · chaos (process-kill a MinIO node — EC tolerates 1). Diagnosed the same v0.8.1-greenfield casualty class as obs (cross-tier CA split + iceberg-pg replication split), fixed by a Greg-authorized cold-rebuild of the Iceberg + Spark envs only (MinIO kept — its EC drives hold 4 cross-tier buckets) + a MinIO IAM key re-sync; 2 live-caught adapter bugs fixed. 27.59 MB AOT; 223/223 tests; ADR-0025. 4 of the 5 non-data tiers done. Preceded by v0.8.3 (2026-06-22) — Phase 0.I: the Observability adapter (ClusterId observability) — the third non-data-tier adapter, managing the Grafana LGTM stack (Prometheus + Loki + Grafana + Tempo + Alertmanager + OTel Collector) across 14 VMs + 2 VRRP VIPs (grafana VIP .184, grafana-db VIP .185; MinIO is the Loki/Tempo S3 backend). Deliberate access divergence caught by diagnosing the live contract first: the obs node TLS leaves are on the tier's OLD CA generation (the tier was offline during the v0.8.1 Vault greenfield) while the build host now trusts the NEW root — so the adapter probes service endpoints over SSH using each node's own ca.crt + reads runtime creds from Vault KV via INexusVaultClient; no managed Prometheus/Grafana/Loki driver (NetArchTest). 8 verbs live-verified GREEN, zero adapter code bugs: status (14 nodes + VIP holders) · health (Prom ready + scrape-targets-up, Alertmanager mesh peers, Loki/Tempo memberlist rings, Grafana database=ok, OTel loopback, Grafana-PG streaming replication, MinIO S3 reachable, both VIPs bound) · topology (14 nodes + 2 VIP pseudo-nodes) · failover = Grafana/Grafana-PG VRRP cutover (--direction grafana proven, RTO≈1.2s) · scale-out = Loki/Tempo memberlist ring add/remove (fixed-HA roles → graceful N/A) · cert-rotate = build-host pki_int/observability-server issue + SSH-push + per-service reload · acl = Grafana users via /api/admin/users · backup = graceful actionable N/A (MinIO erasure-coded + PG repl RPO≈0 + dashboards-as-code + ephemeral Prom TSDB). The verify surfaced three infra divergences (all v0.8.1-Vault-greenfield-while-obs-offline casualties): a tier-wide vault-agent broken-trust state (this drove the SSH-local-curl posture), the Grafana admin password drifted from KV (acl honestly returns HTTP 401 + the reconcile command), and the grafana-pg streaming replication is split (health correctly red); cert-rotate / failover grafana-db / acl grant are implemented but not live-run on the degraded tier (they need a Greg-authorized tier trust re-apply first). 27.59 MB AOT (unchanged from v0.8.2); 194/194 tests; ADR-0024. 3 of the 5 non-data tiers done; the CLI now manages the data tier + 3 non-data tiers (Foundation + Orchestration + Observability). Preceded by v0.8.2 (2026-06-19) — Phase 0.E: the Swarm adapter (ClusterId swarm) — the second non-data-tier adapter and the most reusable: it wires the already-built Consul/Nomad/Portainer clients + ClusterStatusService + FailoverTestService (built v0.1–v0.5) into the full IClusterAdapter surface over the orchestration tier (3 Consul-server/Nomad-server/Swarm-manager nodes + 3 Consul-client/Nomad-client/Swarm-worker/Portainer-agent nodes + a manager-pinned Portainer service); same build-host control-plane posture (Consul/Nomad mgmt tokens from Vault KV, over HTTPS; no managed Docker/Consul/Nomad driver). status/health/topology = the 3-way rollup enriched with docker node ls; failover --direction → consul-leader / nomad-leader (RTO≈2–3s) / swarm-manager (vmrun host-level suspend of the raft-leader VM, RTO≈21s); reversible docker node drain/demote scale-out; consul+nomad snapshot backup round-trip (restore refused on the live cluster); cert-rotate force-reissues the pki_int leaves (the vault-agent pkiCert persists+reuses the cert) then consul ROLLING / nomad PARALLEL big-bang; acl = Consul + Nomad tokens (bootstrap/agent protected); chaos = nexus-chaos.sh on a worker + docker-restart-after-nft. 3 live-caught bugs fixed; cold-rebuild-proven; 27.59 MB AOT; 173/173 tests; ADR-0023. 2 of the 5 non-data tiers done. Preceded by v0.8.1 (2026-06-18) — the first two non-data-tier adapters: VaultAdapter (ClusterId vault) + FoundationAdAdapter (ClusterId foundation-ad) — the CLI now manages the Foundation tier. VaultAdapter drives the 3-node raft Vault HA trust root (vault-1/2/3) + vault-transit (Shamir-seal auto-unseal custodian): control plane = HTTP from the build host with the operator VAULT_TOKEN (root token never ships to a node) + node-local SSH; no managed Vault driver / no shelled vault binary (NetArchTest). All verbs first-try-green — failover = vault operator step-down on the active (standby promotes, RTO≈2.0s); scale-out = stop/start a STANDBY peer; backup = raft snapshot save + non-destructive meta.json inspect (restore refused on the live trust root); cert-rotate = re-issue listener certs from pki_int/vault-server + SIGHUP leader-last; acl = Vault policies+approles; chaos = process-kill a STANDBY — plus a new bespoke recover-ha verb (a new IRecoverableCluster capability = the declarative boot-race recovery: unseal vault-transit → restart vault-1/2/3 → poll unsealed; the only exposed unseal path). Mutating verbs target STANDBYS so the active stays up. FoundationAdAdapter manages the 2-DC nexus.lab AD forest (Windows Server 2025: dc-nexus + dc-nexus-2) over Windows-SSH + the nexus-gateway egress over Linux-SSH; status/health/topology/acl are real, and AD being multi-master, failover/FSMO, DC add/remove, system-state backup, NTDS cert-rotation/chaos return graceful actionable "not applicable" (terraform/out-of-band); health proves both DCs reachable + AD replication result=0 + AD-integrated DNS zones + KDS root key + all 5 FSMO roles + the gateway dnsmasq/nftables/NAT. 1 live-caught bug fixed (the AD-replication probe's explicit -Server arg degraded Get-ADReplicationPartnerMetadata to empty fields — dropped). No new VMs; 27.36 MB AOT; 159/159 tests; ADR-0022. Begins the 5 non-data-tier adapters block (next v0.8.2 Swarm → v0.8.3 Observability → v0.8.4 Lakehouse → v0.8.5 Harbor). Preceded by v0.8.0 (2026-06-18) — the full-fleet roll-up (all 12 data+sharded adapter families sealed + AOT re-validated ≤30 MB, no new adapter). Earlier: v0.7.3 (2026-06-18) — the Citus adapter (ClusterId citus, Phase 0.P): the Citus-sharded PostgreSQL cluster with full Patroni HA — 3 etcd DCS + a coordinator Patroni pair + 2 worker Patroni pairs; PG 17 + Citus 14.1; events hash-distributed on tenant_id (32 shards = 16+16 across the worker groups). Citus = Patroni HA per group + Citus distribution. Operator nexus-cluster-admin (Vault-KV) auto-propagated to the workers + .pgpass so distributed queries run as the operator via the coordinator VIP; topology populates Shards; patronictl switchover failover (RTO≈1.6s, VIP follows the leader); operator COPY round-trip backup of the distributed dataset (800 rows); cert-rotate PG-reload/etcd-restart; acl = PG roles propagated to workers. Adapter first-try-green on every verb; 26.71 MB AOT; 137/137 tests. 12 of 13 adapter families live. Earlier: v0.7.2 (2026-06-17) — the Vitess adapter (ClusterId vitess, Phase 0.O): the Vitess-sharded MySQL cluster — 2 shards ×3 tablets, keyspace commerce hash-vindex on customer_id; hybrid mTLS-control-plane (nexus-vtctldclient) + vtgate :15306 SQL plane; graceful PlannedReparentShard failover (RTO≈0.17s); engine-native file BackupStorage on NFS + xtrabackup (0.O.1; VitessAdapter rewritten); chaos primary-freeze → VTOrc auto-reparent. 26.52 MB AOT; 114/114 tests. Earlier: v0.7.1 (2026-06-16) — the MongoSharded adapter (ClusterId mongo-sharded, Phase 0.N): the first adapter on the 0.7.x line and the genuinely-sharded MongoDB cluster (3 config-server RS + 2 shard RSes ×3 + 2 mongos routers), distinct from the 0.G.2 mongo replica set — two-headed keyFile auth (__system@local for direct mongod RS ops + nexus-sharded-admin@admin through a mongos), topology populates Shards, shard-primary failover RTO≈2.8s, mongodump-through-mongos backup round-trip, cert-rotate now LIVE (0.N.1 wire mTLS — online rotateCertificates on the per-host mongo-sharded-server leaves, no re-election). Preceded by v0.7.0 (2026-06-16) — the Phase 0.G base roll-up (the 0.G exit-gate milestone: the 9 data-tier adapter families sealed + AOT validated ≤30 MB). 10 of 13 adapter families live. Earlier: v0.6.7 (2026-06-15) — the data-tier adapter expansion. Redis (v0.6.0, mTLS-only) + Mongo (v0.6.1, password-auth) + Percona XtraDB Cluster + ProxySQL (v0.6.2, Galera synchronous multi-primary) + PostgreSQL Patroni HA (v0.6.3, streaming replication + etcd DCS + HAProxy VIP) + ClickHouse (v0.6.4, the first sharded + analytics-tier engine — 3 shards × 2 replicas over a ClickHouse Keeper RAFT quorum) + StarRocks (v0.6.5, an MPP MySQL-protocol warehouse — 3 FE BDB-JE metadata quorum + 3 BE tablet nodes) + SQL Server FCI + Always On AG (v0.6.6, the first WINDOWS cluster — two adapters over one vms.yaml cluster: sqlserver WSFC/FCI + sqlserver-ag AG/Listener; Windows-SSH + sqlcmd, no managed Microsoft.Data.SqlClient; FCI Move-ClusterGroup failover RTO≈4.5s + AG ALTER AVAILABILITY GROUP FAILOVER RTO≈8.2s; Listener strict-TLS; one shared FCI cert + single cluster checkpoint; manual-seed AG scale-out) + Kafka (v0.6.7, the last data tier — kafka-east + kafka-west KRaft clusters + the kafka-ecosystem observe adapter; mTLS-only like Redis, no managed driver; controller-leader failover RTO≈4.5s; enabled the KRaft StandardAuthorizer so acl enforces; zero live bugs) ship all data-tier verbs green — status/health/topology/failover (StarRocks FE leader re-election RTO≈1.5s; ClickHouse Keeper re-election RTO≈1.1s; SQL Server FCI Move-ClusterGroup RTO≈4.5s + AG failover RTO≈8.2s)/cert-rotate/acl/backup (StarRocks genuine async BACKUP/RESTORE SNAPSHOT to the file:// NFS repo; SQL Server COPY_ONLY RESTORE-WITH-MOVE round-trip)/scale-out/chaos. The Vault-KV operator-credential model (the nexus-cluster-admin password lives only in Vault KV, fetched at runtime via an optional INexusVaultClient) carries across every password-auth adapter; 25.95 MB under the ≤30 MB gate. 5 adapters follow. 0.G.3.5 architectural refactor (per-cluster Terraform state + per-engine Packer template — the rule born from 0.G.3's 16-transient stall) canonized; shrinks per-cluster iteration from ~30 min monolithic → ~5-10 min. 0.G.4 (Patroni PG 17 HA + etcd 3.5 DCS + HAProxy 3 HA pair + VRRP VIP .60, 8 VMs) ✅ CLOSED 2026-05-19 end-to-end — smoke gate 152/152 ALL GREEN; 18 transients surfaced + all root-caused + permanently fixed in source. 0.G.7 (SQL Server FCI + Always On AG on Windows Server 2025, 4 nodes) ✅ LIVE-RATIFIED 2026-05-22 — smoke smoke-0.G.7.ps1 ALL GREEN 56/56. Full canon hybrid FCI+AG: a 2-node WSFC Failover Cluster Instance (sqlfci @ .70.16) sharing an iSCSI LUN from nexus-gateway (per ADR-0026) + 2 async AG replicas; AG Listener (sql-ag-listener @ .70.17) is the LB-tier HA primitive (per ADR-0025) — a remote domain client sqlcmd -E -N (Encrypt + strict cert-chain validate) reaches the primary across the listener; SQL service identity = gmsa-sql-engine$ on the FCI (first real GMSA consumer; 0.D.5 scaffolded the infrastructure); AG endpoint authentication cert-based per ADR-0027. First Windows-fleet data cluster + first iSCSI shim; 40+ ratification transients all fixed in source. OLTP tier SEALED — 5/5 clusters cold-rebuild proven (redis + mongo + percona + patroni + sqlserver-fci-ag). The analytics tier (ClickHouse 3×2 + 3-node Keeper, StarRocks 3 FE + 3 BE — nexus-infra-analytics v0.1.0) and the lakehouse + registry tier are live + cold-rebuild-proven: nexus-infra-lakehouse (MinIO distributed-EC object store + Iceberg/Nessie REST on a dedicated PG HA pair + Apache Spark HA with a ZooKeeper-elected master pair), and nexus-infra-registry — a highly-available Harbor container registry (2 stateless app nodes + dedicated PG/Redis HA datastore + VRRP VIP; image blobs in MinIO S3; Trivy scanning + cosign signing; Vault OIDC SSO → AD; ADR-0036). Phase 0.L.5 StarRocks shared-data SEALED + analytics v0.2.0 released (2026-05-26) — second StarRocks cluster live + cold-rebuild-proven in nexus-infra-analytics (3 FE BDB-JE quorum + 2 stateless Compute Nodes, run_mode=shared_data, internal cloud-native tables in a MinIO storage volume s3://starrocks/ with a dedicated nexus-starrocks-app service account + scoped starrocks-tenant policy; ADR-0037 amends ADR-0030; smoke-0.L.5.ps1 69/69 GREEN with CN-loss chaos default-on, proving the shared-data HA property — any CN serves any query from shared MinIO storage). Fleet 105 VMs

• 5 VIPs built/cold-rebuild-proven. 0.L.6 close-out COMPLETE 2026-05-26 — 3 tags shipped (nexus-infra-lakehouse v0.1.0 · nexus-infra-registry v0.1.0 · nexus-infra-analytics v0.2.0). Phase 0.I observability COMPLETE 2026-05-28 (ADR-0038) -- nexus-infra-observability v0.1.0 tagged. All 7 sub-phases SEALED (live-ratified + cold-rebuild-proven). Grafana LGTM stack fully up: Prom HA + Alertmanager mesh (0.I.1) · Loki SSD on MinIO (0.I.2) · Tempo scalable on MinIO (0.I.3) · Grafana HA + Grafana PG HA + 2 VRRP VIPs grafana.nexus.lab .184 + grafana-db.nexus.lab .185 (0.I.4) · OTel Collector pair behind RR DNS otel.nexus.lab (0.I.5) · Vector log shipper added to shared nexus_observability ansible role + live retrofit via scripts/fleet-vector-rollout.ps1; every fleet Linux node ships journald + /var/log/ to Loki HA tagged fleet=nexusplatform (0.I.6)* · 3-layer canon sweep + tag (0.I.7). 39 transients permanently fixed in source. 14 obs VMs + 2 VIPs (fleet 107 + 5 VIPs cold-rebuild-proven). Phase 0.N MongoDB sharded cluster SEALED 2026-05-30 (ADR-0040) — live-ratified + cold-rebuild-proven (smoke-0.N.ps1 50/50 GREEN both times). 11 VMs: 3 config-server RS + 2 shard RSes (3 each) + 2 stateless mongos routers — the document-store sharding showcase (chunks span both shards; mongos routes keyed lookups) alongside the 0.G.2 replica-set (replication) showcase. keyFile member auth; 10 transients fixed in source. OLTP tier now 6/6 cold-rebuild-proven. 0.N.1 (2026-07-10) added Vault-PKI wire mTLS — net.tls.mode requireTLS + per-host mongo-sharded-server leaves + online rotateCertificates, parity with the 0.G.2 RS (clusterAuthMode stays keyFile for member auth). (Prior: Phase 0.M foundation HA COMPLETE 2026-05-28, ADR-0039 — 2nd AD DC dc-nexus-2 closes the last single-DC SPOF.) Phase 0.O nexus-infra-vitess SEALED 2026-06-03 (ADR-0041) — live-ratified + cold-rebuild-proven (smoke-0.O.ps1 71/71 GREEN both times); tagged v0.1.0. 12 VMs on tier 07-vitess: 3-node etcd 3.5.16 topo + a control node (vtctld + VTOrc) + 2 stateless vtgate routers (RR DNS vtgate.nexus.lab, MySQL :15306) + 2 shards × 3 tablets (vttablet + Percona Server 8.4 LTS). Keyspace commerce, 2 shards (-80/80-) on a hash vindex — one vtgate insert splits 100 rows 53/47 across both shards; VTOrc auto-reparents a shard (~15s) on PRIMARY kill. Vitess v24.0.1; full Vault-PKI mTLS on every gRPC channel + the mysqld wire + the vtgate MySQL listener. The relational (MySQL) sharding showcase, distinct from PXC/Galera replication (0.G.3). Phase 0.P nexus-infra-citus SEALED 2026-06-03 (ADR-0042) — live-ratified + cold-rebuild-proven (smoke-0.P.ps1 69/69 GREEN both times incl worker Patroni failover); tagged v0.1.0. 9 VMs + 3 VRRP VIPs on tier 08-citus: 3-node etcd 3.5 DCS + a coordinator Patroni pair (VIP coord.citus.nexus.lab) holding the pg_dist_* catalog + 2 worker Patroni pairs (VIPs worker{1,2}.citus.nexus.lab) holding the shards. **PostgreSQL 17
• Citus 14.x**; a distributed events table (32 shards) spreads across both worker groups, with reference + colocated tables. Every node-group is a 2-node Patroni cluster over the shared etcd DCS, fronted by a keepalived VRRP VIP that follows the Patroni leader (workers registered in pg_dist_node by VIP, so a failover needs no metadata rewrite). Full Vault-PKI mTLS on the PG wire + etcd + Patroni REST. The relational (PostgreSQL) sharding showcase with full Patroni HA, distinct from Patroni streaming replication (0.G.4). Fleet 140 VMs + 8 VIPs cold-rebuild-proven. After 0.P came the hardening pass — 0.N.1 sharded-Mongo wire mTLS + 0.O.1 Vitess engine-native backup + the platform CA rollover (2026-07-04/05) + nexus-cli reaching 100% (v0.8.12, 13 GAPs closed) — all now DONE. Phase 0 is COMPLETE; the first application project — dataflow-studio — has shipped v0.1.0 (2026-07-25).

●  .NET 10 / C# 13 · ASP.NET Core · Blazor Server · gRPC · Native AOT
●  Kafka (KRaft, Streams, ksqlDB, Debezium, MirrorMaker 2) · Schema Registry
●  SQL Server Always On · Percona MySQL PXC · PostgreSQL Patroni · MongoDB RS · Vitess (MySQL sharding)
●  StarRocks · ClickHouse · Redis Cluster · Kimball modelling · SCD2
●  ML.NET · ONNX · PyTorch→ONNX · Semantic Kernel · Ollama · HuggingFace
●  Docker Swarm · HashiCorp Nomad · Consul · Vault · Kubernetes · Terraform · Packer
●  OpenTelemetry · Prometheus HA · Grafana HA (VRRP VIP) · Loki · Tempo · Alertmanager
●  Windows Forms · WPF · WinUI 3 · MAUI · MVVM · ReactiveUI

Full matrix with project mapping: portfolio-index/PORTFOLIO.md.

• Email: gzapas@gmail.com
• LinkedIn: https://www.linkedin.com/in/grigoris-zapantis-1a0638b/
• Upwork: https://www.upwork.com/freelancers/~01de3f0552684544ee
• CV: <link once hosted>