Senior Software Engineer at Allegro

Built an event-driven pricing/checkout data plane with Kafka topics partitioned by merchant+SKU and idempotent Debezium CDC from OLTP into ClickHouse for sub-second analytics. Implemented real-time materialized views (Parquet on S3 + Iceberg) refreshed via Airflow with incremental upserts; reduced dashboard drill-down query time from 2.8 s to 320 ms. Led RAG-ready catalog enrichment using offline embeddings stored in pgvector with HNSW for semantic substitutes/upsells, increasing attach-rate by 3.1% in AB tests. Established data quality SLAs with Great Expectations and data contracts, preventing 4 broken deploys and reducing downstream incidents by 40%. Optimized storage and scans with column pruning, ZSTD, Iceberg compaction, saving ~$6.2k/month. Deployed anomaly detection on checkout/payout flows; identified regressions pre-release and rolled back within 15 minutes. Implemented end-to-end tracing (OTel) across browser RUM → API → Kafka → Airflow → ClickHouse; improved debuggi

Senior Software Engineer at Artisan AI

Contributed to a Search & RAG platform with hybrid retrieval (BM25 + pgvector) and Redis Bloom prefilters; achieved F1 +13% and reduced tail latency for multilingual corpora. Bootstrapped Feast over ClickHouse/Postgres for offline/online consistency; sped up label joins by 9× and eliminated leakage bugs. Managed ML lifecycle with MLflow registry and staging/prod canaries via KServe; automated shadow traffic and AB testing with p-value gates; rollback <5 min. Implemented batch (Airflow) and streaming (Kafka) inference with autoscaled consumers achieving p95 ~180 ms at 2.1k RPS while maintaining GPU utilization >75%. Enforced guardrails (prompt sanitizers, PII filters, grounding checks) reducing hallucinations by 28% and policy violations by 62%; documented red-team playbooks. Built evaluation harness (BLEU/ROUGE) with human-in-the-loop UI (React) cutting weekly eval cycles by 50%. Ensured vector hygiene with periodic HNSW rebuilds, dedupe, and drift monitoring; reduced false positives

DevOps Engineer at AioCare

Rolled out a Kubernetes platform (3 environments, 8 node pools) using Terraform + Helm + Argo CD, reducing deploy time from 45 minutes to 7 minutes and enabling ~20 production releases per week. Implemented progressive delivery (blue/green, canary up to 100%) via GitHub Actions → Argo Rollouts, lowering failed deployments by ~72% and MTTR from 38 to 11 minutes. Introduced SLOs (p95 API < 250 ms, 1%/30d error budget) with burn-rate alerts, achieving 99.95% API uptime over 12 months and MTTD reduction to 3 minutes. Optimized cloud spend by 28% YoY through HPA/VPA, spot instances, storage lifecycle policies, and right-sizing; reduced environment spin-up from days to hours. Shipped end-to-end observability (OpenTelemetry, Prometheus, Loki/ELK) with golden dashboards, reducing unknown root-cause postmortems by ~60% and after-hours paging by ~50%. Hardened SDLC for medical data (SBOM, image signing, OPA/Gatekeeper, Vault/KMS) reducing CVEs by >90% in two quarters. Delivered DR runbooks and

Software Engineer at Spyrosoft

Designed, developed, and maintained enterprise and consumer-facing apps for international clients (e.g., fintech dashboard for BNP Paribas, e-commerce platform for a retail client, healthcare management tool for a telemedicine startup). Delivered scalable front-end solutions (React/Angular/Vue) and high-performance backends (FastAPI, Django, Java, Go), ensuring secure integration and strong user experience.

Senior Software Engineer at Artisan AI

Contributed to a Search & RAG platform with hybrid retrieval (BM25 + pgvector) and Redis Bloom prefilters, achieving +13% F1 and ~260 ms tail latency on multilingual corpora. Built a Feast-based feature store over ClickHouse/Postgres for offline/online consistency; fast label joins reduced leakage bugs to zero. Managed ML lifecycle with MLflow registry and KServe canaries; automated shadow traffic and A/B with p-value gates and rollback in under 5 minutes. Implemented batch (Airflow) and streaming (Kafka) inference; autoscaled consumers to meet p95 ~180 ms at 2.1k rps with GPU utilization >75%. Enforced guardrails (prompt sanitizers, PII filters, grounding checks) reducing hallucinations by 28% and policy violations by 62%; documented red-team playbooks. Built an evaluation harness (BLEU/ROUGE) with human-in-the-loop labeling UI, cutting weekly eval cycle by 50%. Maintained vector hygiene with periodic HNSW rebuilds and drift monitoring, reducing false positives by 18%. Aligned data me

DevOps Engineer at AioCare

Rolled out a Kubernetes platform (3 environments, 8 node pools) using Terraform, Helm, and Argo CD, cutting deployment time from 45 minutes to 7 minutes (-84%) and enabling ~20 prod releases per week. Implemented progressive delivery with blue/green and canary (5%→25%→100%) via GitHub Actions to Argo Rollouts, reducing failed deployments by ~72% and MTTR from 38 minutes to 11 minutes. Introduced SLOs (p95 API < 250 ms, 1% error budget/30d) improving API uptime to 99.95% over 12 months. Optimized cloud spend by 28% YoY using HPA/VPA, spot nodes, lifecycle policies, and right-sizing. Shipped end-to-end observability (OpenTelemetry, Prometheus, Loki/ELK) reducing unknown root-cause postmortems by ~60% and after-hours pages by ~50%. Hardened SDLC for medical data with SBOM, image signing, OPA/Gatekeeper, and Vault/KMS; reduced critical CVEs by >90% in two quarters. Delivered DR runbooks achieving RPO ≤ 5 minutes and RTO ≤ 30 minutes with cross-region backups. DR drill results docum