Danila Yurkov

Backend & AI Engineer

Python · TypeScript · Java · Go · Node.js · FastAPI · NestJS · PostgreSQL · LLM tooling · MCP · RAG

Portfolio · All repositories

I build backend and AI systems: APIs, data layers, integrations, AI agents, retrieval and context infrastructure, realtime services, and internal automation.

I use Claude Code and Codex as part of my engineering process, from breaking down requirements and designing architecture to review, testing, and deployment. My main interests are agentic systems, context engineering, MCP, RAG, and tooling that lets AI agents work with large codebases.

I'm open to Backend / Python / AI Developer roles (Junior+ to Middle-) and strong engineering internships.

A local retrieval and context layer for Claude Code, Codex, OpenCode and MCP clients. It indexes a repository, extracts symbols and dependencies with Tree-sitter, and gives an AI agent relevant context with exact file:line references.

Python · SQLite / FTS5 · Tree-sitter · MCP · Hybrid / vector retrieval

• hybrid retrieval with opt-in vector search;
• symbol, call, import and inheritance graph;
• impact analysis and diff-aware impact analysis;
• token-budgeted retrieval and skeletonized context;
• CLI, skills/plugins and an MCP server;
• runs locally with no network access by default.

Measured benchmark: on one real ~55k LOC Java codebase, Recall@3 was 70% vs 40% for rg + window, with roughly 13x less answer context. This applies only to the published benchmark and is not claimed as a general result.

Coordinates several AI coding-agent sessions working in one repository.

Python · SQLite · Claude Code skills / hooks

• shared task board with dependencies;
• file and resource locks with TTL;
• messaging and shared state between agents;
• activity log and architecture decision records;
• Claude Code hooks that prevent conflicting edits;
• live operator console for parallel agent work.

A Fabric/NeoForge mod that lets AI agents control and observe a running Minecraft: Java Edition game over MCP (1.21.1+).

TypeScript · MCP · Fabric · NeoForge

Event-driven backend for event processing and analytics, focused on concurrency and reliability.

Go · Kafka · ClickHouse · PostgreSQL · Redis · Docker Compose

• bounded worker pool with backpressure;
• retries and a dead-letter topic;
• contiguous per-partition offset tracking;
• PostgreSQL and ClickHouse migrations;
• health checks and Prometheus metrics;
• integration tests with Testcontainers.

Structured log ingestion and search service with a durable indexing pipeline.

Python · FastAPI · PostgreSQL · Elasticsearch · Redis · Docker · GitLab CI

• PostgreSQL-backed jobs with leases, retries and deterministic document IDs;
• typed full-text, exact-filter and facet search;
• health and readiness endpoints, Prometheus metrics, integration tests.

A local RAG workbench with a transparent retrieval-to-answer pipeline and an evidence trail for every result.

Python · FastAPI · React · Qdrant · FastEmbed · Docker Compose

• deterministic chunking and document IDs;
• Qdrant retrieval with source filters, score threshold and bounded context;
• extractive mode by default, optional OpenAI-compatible provider;
• citation validation, refusal on insufficient evidence, prompt-injection protection;
• golden evals for recall, citations, refusal correctness and injection regressions;
• React evidence workbench, health/readiness, metrics, CI and Compose smoke path.

Local-first pipeline for ordered device signals: UDP/TCP ingress, Kafka transport, a Spring processor, PostgreSQL persistence and a live dashboard.

NestJS · TypeScript · Kafka · Java · Spring Boot · PostgreSQL · WebSocket · React

• bounded UDP/TCP ingress with validation and a framed protocol;
• idempotent processing by event_id with a per-source sequence watermark;
• late-event classification, DLQ metadata, readiness and metrics;
• React dashboard with REST snapshot, live updates, and explicit loading/empty/error states;
• Testcontainers integration tests, Compose smoke client, CI and secret scan.

An out-of-home advertising management system built from a real-world functional spec.

Next.js 16 · React · TypeScript · PostgreSQL · Prisma · Auth.js · Vitest · Playwright

• CLIENT / ADMIN roles and protected routes;
• REST API, CRUD and server-side business logic;
• map, filtering, calendar scenarios and calculations;
• CSV/XLSX import with idempotent upsert, and Excel export;
• unit and E2E tests.

Some source code is private. I'm happy to walk through the architecture and show relevant parts in a technical interview.

An agentic coding platform covering the path from idea and requirements to architecture, backlog, AI-agent orchestration, and deployment infrastructure.

NestJS · TypeScript · Next.js · React · Prisma · PostgreSQL · Redis · BullMQ · WebSocket · Docker

My scope: project architecture and the backend/full-stack implementation, including an AI-agent framework with provider abstraction, API, queues, realtime, integrations, deployment, and the infrastructure layer.

A web platform with accounts, a shop, admin tooling, bots and external integrations, plus the Minecraft server plugins behind a live multiplayer network.

Next.js · React · TypeScript · PostgreSQL · Prisma · Java · Docker Compose · GitHub Actions

Hands-on production operations: database migrations, CI/CD, health checks, integrations, incident fixes, and deployment on Linux/VDS.

Backend: Python · TypeScript · Node.js · NestJS · FastAPI · REST · WebSocket · Java · Go
Frontend: React · Next.js · TypeScript · SSR
Data: PostgreSQL · MySQL · SQLite · Prisma · SQLAlchemy · Redis · ClickHouse
AI / LLM: MCP · RAG / retrieval · AI agents · context engineering · LLM integrations
Infrastructure: Docker · Docker Compose · Linux · Nginx / Caddy · GitHub Actions · CI/CD
Testing: pytest · Vitest · Playwright · integration / E2E testing
Other: OpenAPI · JWT / sessions · OAuth · BullMQ · Kafka · S3 / MinIO

• retrieval and context pipelines for LLMs;
• MCP and integrating AI agents with tools;
• backend architecture in Python / TypeScript;
• REST API and data model design;
• async work, queues and realtime;
• Docker, CI/CD and production deployment;
• AI-assisted development: decompose → implement → review → test → security checks.