Evgenii Perminov

Offer Inspector – SaaS, in production

• Built and launched Offer Inspector, a full-stack salary comparison SaaS using Next.js, React, NestJS, PostgreSQL and Go workers, covering authentication, billing, tax logic, localization and production infrastructure.
• Designed an AST-based tax engine for salary calculations across Estonia and Poland, supporting multiple legal forms, net-to-gross calculations, employer costs, paid leave and effective salary comparisons.
• Set up GitOps-based deployment on a self-hosted k3s cluster with Argo CD, blue-green releases, automated database migrations and encrypted secrets.
• Added application and infrastructure observability using OpenTelemetry, Prometheus, Grafana, Loki and Tempo, including database metrics, tracing, dashboards and Slack alerts.

Embeddorium – open-source retrieval-engineering workbench

github.com/PerminovEugene/Embeddorium

• Built a distributed ingestion pipeline with Dramatiq, RabbitMQ, PostgreSQL outbox events, idempotent jobs and bounded retries.
• Separated canonical datasets from indexing pipelines, allowing the same corpus to be tested with different chunkers and embedding models while keeping experiment configurations reproducible.
• Implemented dense, BM25 and hybrid retrieval with RRF, DBSF, weighted score fusion and optional cross-encoder reranking.
• Built a chunker-independent evaluation system using document character spans, with Recall@K, MRR, nDCG, Hit Rate and Passage Coverage metrics.
• Added auto-discovered plugins for parsers, chunkers, fetchers, filters and model providers.
• Built an MCP server exposing provider, dataset preparation, indexing and staged search workflows to AI agents.

Stack: Python, FastAPI, Dramatiq, RabbitMQ, PostgreSQL, Qdrant, Ollama, OpenAI API, LangChain, LangGraph, MCP, React, Vite, Docker Compose