Denis S.

I build the backend that AI features run on: LLM agents, RAG over your own data, evals, and the production plumbing around them. Behind it: 12+ years of high-load Go and Python backends for fintech, payments, trading and iGaming. 17,000+ Upwork hours, 100% Job Success. Most AI projects do not fail on the model. They fail on the system around it: slow retrieval, no streaming, no retries, no evals, costs out of control, no way to see what the agent did. That system is my strength. AI AGENTS & LLM BACKENDS - LLM agents and tool calling, orchestrated as durable workflows (Temporal) with retries, timeouts and human approval steps - RAG: embeddings, vector search (pgvector and others), chunking and retrieval tuning over private data - Multi-provider LLM backends: Claude, OpenAI, Gemini and local models through Ollama - Streaming chat and assistants with request cancellation - MCP servers that connect your APIs and data to Claude and other agents - AI features added to an existing product, not a demo next to it EVALS, OBSERVABILITY & COST - Eval suites in CI: regression tests for prompts, retrieval quality and agent behaviour - Tracing, token and cost metrics, rate limits and fallbacks between providers - Guardrails for money and data: idempotency, audit logs, access control CLAUDE CODE, CODEX & AI-ASSISTED DEVELOPMENT - Claude Code is my daily driver. I build skills, hooks, MCP servers and agent workflows - I set up AI-assisted development for teams: shared conventions, review gates, task automation - Same tooling for Claude Code, Codex and OpenCode - Anthropic "Claude Code in Action" certificate HIGH-LOAD BACKEND, PAYMENTS & PLATFORM - Lead Backend on two iGaming platforms: core platform, payment gateway, public API, game logic, CI/CD - Scaled a casino platform to 5,000+ concurrent users at 1,000+ RPS and moved it from IBM to AWS - Payments, wallets, ledgers, idempotency and reconciliation - Real-time systems: WebSockets, gRPC, Kafka, event-driven pipelines - Kubernetes, Helm, Terraform, GitHub Actions on AWS and GCP - Full-stack when needed: React/Next.js and TypeScript front ends on top of the backend, delivered end-to-end with AI-assisted development TRADING TOOLS & AUTOMATION - Trading automation: signal-to-order execution over exchange and broker APIs - Backtesting runners, market data ingestion, tick and OHLCV storage - Low-latency services and real-time market data pipelines - I am the engineer, not the quant: you bring the strategy, I build the system that runs it EDGE AI & DEVICES - On-device inference on Raspberry Pi 5 and NVIDIA Jetson Orin Nano - Local LLM and embedding models tuned for small-memory boards - Offline-first device stacks with cloud sync and backup STACK Go, Python, TypeScript, Kotlin, Java, C React, Next.js, Node.js PostgreSQL (pgvector), Redis, MongoDB, vector search Temporal, Kafka, RabbitMQ, PubSub, gRPC Docker, Kubernetes, Helm, Terraform, AWS, GCP Claude and OpenAI APIs, Ollama, LangChain Go, MCP TRACK RECORD - Now: Lead Backend at an iGaming company (Go, Postgres, Redis, Temporal, Kubernetes) and an AI-powered consumer device: LLM chat, face recognition pipeline, protected backups - B2B credit card product: real-time scoring, risk tooling, public API - Neobank QA API and ClearBank simulator, wired into CI/CD - Co-founded a software company: desktop apps, CLI tools, ML for image recognition HOW I WORK Simple systems that stay simple as they grow. Clear contracts, real monitoring, direct communication, no over-selling. Available part-time for contracts, consulting and short projects. Tell me what you are building.