Top Sites>Roman Matveev - Python backend engineer

No stored offer samples.

No stored action samples.

No stored product samples.

# Roman Matveev — extended agent context
> Canonical locale for this file: en
> Person: Roman Matveev — Backend Engineer (Python, integrations, applied AI) — Kazan, Russia · Remote · UTC+3
> Preferred markdown entry: https://ramenm.com/en/index.md

## Locale indexes
- Russian llms.txt: https://ramenm.com/ru/llms.txt
- Russian llms-full.txt: https://ramenm.com/ru/llms-full.txt
- English llms.txt: https://ramenm.com/en/llms.txt
- English llms-full.txt: https://ramenm.com/en/llms-full.txt
- Simplified Chinese llms.txt: https://ramenm.com/zh/llms.txt
- Simplified Chinese llms-full.txt: https://ramenm.com/zh/llms-full.txt

## Russian surface
- Home HTML: https://ramenm.com/ru
- Home Markdown: https://ramenm.com/ru/index.md
- Chat HTML: https://ramenm.com/ru/chat
- Chat Markdown: https://ramenm.com/ru/chat/index.md
- Russian llms.txt: https://ramenm.com/ru/llms.txt
- Russian llms-full.txt: https://ramenm.com/ru/llms-full.txt

## English surface
- Home HTML: https://ramenm.com/en
- Home Markdown: https://ramenm.com/en/index.md
- Chat HTML: https://ramenm.com/en/chat
- Chat Markdown: https://ramenm.com/en/chat/index.md
- English llms.txt: https://ramenm.com/en/llms.txt
- English llms-full.txt: https://ramenm.com/en/llms-full.txt

## Simplified Chinese surface
- Home HTML: https://ramenm.com/zh
- Home Markdown: https://ramenm.com/zh/index.md
- Chat HTML: https://ramenm.com/zh/chat
- Chat Markdown: https://ramenm.com/zh/chat/index.md
- Simplified Chinese llms.txt: https://ramenm.com/zh/llms.txt
- Simplified Chinese llms-full.txt: https://ramenm.com/zh/llms-full.txt

## Site summary
I build backend systems and integrations that still work after launch.

I take rough briefs and turn them into services: APIs, integrations, automation, and LLM features.

Open to remote or hybrid full-time roles, plus selective contract work on difficult integrations.

## Approach
I do my best work in messy environments: unstable external APIs, awkward workflows, and AI layers on top of real operations. The goal is to make delivery predictable and supportable.

### What I like building
- Python backends with clear API contracts
- Integrations and automation for unreliable external systems
- Applied AI features (LLM APIs, RAG, tool calling) that fit the workflow
- Observability, debugging, and support after launch

### Principles
- **Constraints first** — I start with external systems, data shape, operational risk, and support cost.
- **Simple systems last longer** — I prefer systems where data flow and failure boundaries are obvious.
- **Production is the real test** — A solution is ready when people can monitor it, debug it, and extend it safely.
- **Good delivery compounds** — Good backend work removes manual steps and makes the next release cheaper.

## Proof points
- **Commercial experience**: 5+ years — Production backend, integrations, automation, and applied AI work.
- **LLM product work**: 500+ models — Built a live catalog for model research and comparison.
- **Engineering focus**: Reliability + speed — I build systems teams can scale and keep running.
- **Delivery scope**: From brief to launch — From task framing and API contracts to rollout and stabilization.

## Featured projects
Case studies told as problem -> decision -> outcome.

### nnzen model catalog
- URL: https://ramenm.com/en/projects/llm-models-hub
- Markdown: https://ramenm.com/en/projects/llm-models-hub/index.md
- Role: Founder / Backend Engineer
- Period: 2025-2026
- Team: Solo
- Summary: A live catalog with 500+ model cards that makes model research less scattered.
- Problem: Model data was spread across different sources, so pricing, context size, limits, and quality signals had to be checked by hand.
- Solution: Built a FastAPI backend that pulls in OpenRouter data, normalizes model cards, adds ranking context, and exposes search and filters.
- Impact: Model choice went from tab-hopping to one place.
- Stack: Python, FastAPI, LLM APIs, RAG, Vector DB / pgvector, Tool calling
- Metrics:
  - Catalog: 500+ model records
  - Data flow: Auto-enriched
  - Format: Live production tool
- Highlights:
  - Designed the ingestion pipeline for collection, normalization, and catalog updates.
  - Built a unified comparison surface across model metadata and ranking context.
  - Kept the backend extensible for new sources, filters, and comparison layers.
- Lessons:
  - Even data-heavy tools benefit massively from strong information architecture.
  - Faster decision-making often matters more than infinitely detailed analysis.
- Links:
  - [Open nnzen.com](https://nnzen.com)

### Custom Agent Core with MCP
- URL: https://ramenm.com/en/projects/enhanced-mcp-agent
- Markdown: https://ramenm.com/en/projects/enhanced-mcp-agent/index.md
- Role: AI Tooling Engineer
- Period: 2025
- Team: Solo
- Summary: LLM core with a plugin execution layer for a developer assistant: hot reload, tool chains, and explicit context handoff.
- Problem: Once an assistant gets more capable, plugins become hard to evolve and orchestration gets brittle.
- Solution: Built an EnhancedMCP core: FastAPI server, plugin lifecycle management, hot reload, cascading tool calls, and a CLI client with execution history.
- Impact: The result is a reusable core that can support new workflows and tools without a full rewrite.
- Stack: Python, FastAPI, MCP, Tool calling, LLM APIs, Docker / Docker Compose, TypeScript
- Metrics:
  - Plugin model: Hot-reload
  - Execution: Context handover
  - Surface: CLI + API
- Highlights:
  - Implemented plugin hot reload without restarting the core process.
  - Built explicit context handoff between chained tools.
  - Separated core runtime responsibilities from plugin responsibilities to keep the platform maintainable.
- Lessons:
  - Architecture matters more than model count in AI tooling.
  - Developer trust comes from predictable execution behavior.

### Trading Automation for an In-Game Marketplace
- URL: https://ramenm.com/en/projects/marketplace-trading-bot
- Markdown: https://ramenm.com/en/projects/marketplace-trading-bot/index.md
- Role: Backend / Automation Engineer
- Period: 2024
- Team: Solo
- Summary: Automation for a constrained external marketplace with strategy logic, execution control, and logging.
- Problem: The platform was unstable enough that naive scripts failed quickly.
- Solution: Designed the API layer, strategy controls, and execution loop from scratch, based on reverse engineering and the platform's actual behavior.
- Impact: The system could keep running under platform changes instead of falling apart.
- Stack: Python, FastAPI, REST APIs / Webhooks, Reverse engineering, Docker / Docker Compose
- Metrics:
  - Workflow: End-to-end
  - Control: Strategy layer
  - Environment: Constrained external platform
- Highlights:
  - Designed API interaction, execution control, and logging as one operational loop.
  - Adapted automation behavior to real platform constraints instead of ideal assumptions.
  - Kept the system supportable under frequent external changes.
- Lessons:
  - Good automation starts with a stable service layer, not with UI macros.
  - Reverse engineering is only useful when it becomes a maintainable interface.

### Resilient Data Collection Tooling
- URL: https://ramenm.com/en/projects/resilient-data-collection
- Markdown: https://ramenm.com/en/projects/resilient-data-collection/index.md
- Role: R&D Data Collection Engineer
- Period: 2021-2023
- Team: Product team
- Summary: Data extraction tooling for a changing web environment with persistent anti-bot friction.
- Problem: Standard collection approaches kept breaking because of page drift, client-side logic, and defensive mechanisms.
- Solution: Worked at the intersection of Python, HTTP, and JavaScript reverse engineering: traced request flows, adjusted extraction logic, and hardened the pipeline.
- Impact: Less firefighting, more predictable extraction.
- Stack: Python, Web scraping, Reverse engineering, Playwright, ClickHouse
- Metrics:
  - Context: High-friction web
  - Focus: Resilience
  - Approach: HTTP + JS analysis
- Highlights:
  - Reworked unstable request flows into repeatable extraction logic.
  - Improved resilience to anti-bot changes and page-structure drift.
  - Balanced delivery speed with reliability under constant external change.
- Lessons:
  - Data collection is an infrastructure problem as much as an extraction problem.
  - Stability comes from understanding the request model, not from brute-force retries.

## Experience
Experience is organized as problem, decision, and result: what was risky, what choice I made, and what changed after release.

### Freelance — Backend / Integration Engineer
- Period: Feb 2024 — Present
- Mode: Contract · delivery ownership
- Summary: Build backend, integration, and automation systems for real operating workflows across external APIs, process automation, and AI-assisted features.
- Impact:
  - Turn ambiguous requirements into maintainable services with explicit API contracts and support boundaries.
  - Design integrations for unreliable external systems with robust error handling, retries, logging, and observability.
  - Use reusable modules and integration patterns to shorten delivery time across new workflows.
- Stack: Python, FastAPI, REST APIs / Webhooks, Docker / Docker Compose, LLM APIs, RAG, Tool calling, Reverse engineering

### Independent projects / R&D — Independent R&D Engineer
- Period: Mar 2023 — Jan 2024
- Mode: Self-directed research
- Summary: Built focused backend and AI R&D between contracts, validating architecture patterns before using them in production-facing work.
- Impact:
  - Validated RAG and tool-calling patterns on working prototypes before client use.
  - Shifted from one-off scripts to reusable service components with clear interfaces.
  - Built a practical base that later fed applied AI and AI tooling projects.
- Stack: Python, LLM APIs, RAG, MCP, Tool calling

### Bright Data — R&D Data Collection Engineer
- Period: May 2021 — Feb 2023
- Mode: Full-time
- Summary: Built and maintained data-collection tooling in a changing web environment: HTTP/JavaScript analysis, resilient request flows, and fast adaptation.
- Impact:
  - Analyzed unstable request chains and turned them into more resilient extraction logic.
  - Maintained extraction quality as anti-bot controls and page structure shifted.
  - Worked at the intersection of reverse engineering, reliability, and delivery speed.
- Stack: Python, Reverse engineering, Web scraping, Playwright

### Freelance — Python Developer
- Period: Sep 2019 — May 2021
- Mode: Contract
- Summary: Built Python tools, parsers, and automation for internal workflows with messy inputs and changing requirements.
- Impact:
  - Automated repetitive operations and reduced manual effort in data-heavy processes.
  - Built service utilities and parsers for unstable integrations.
  - Shipped practical Python tools quickly and stabilized them for ongoing use.
- Stack: Python, Web scraping, pandas / ETL

## Ask about the work behind the portfolio
If you'd rather ask directly, the assistant answers from public case studies and the decisions behind them.

### Suggested prompts
- How do you approach LLM integrations in production?
- Which project best shows your decision-making?
- What backend work do you usually own from start to finish?
- How do you keep AI tooling reliable?

### Trust and usage notes
- Grounded in public case studies and the current backend context.
- If Turnstile is enabled, verification is required before sending a message.
- Best for project details, architecture, and working style. Don't share sensitive or private client data.

## Have a messy brief? Let's turn it into a working system.
I'm a good fit when the brief is fuzzy, the systems are real, and the result has to survive production.

Remote or hybrid. Open to full-time roles and selective contract work.

### Contact links
- [Email](mailto:ramen44@yandex.ru)
- [Telegram](https://t.me/ramenm44)
- [VK](https://vk.ru/nyashpy)
- [GitHub](https://github.com/Ramenm)
- [nnzen.com](https://nnzen.com)

## Alternate locale entry points
- English markdown: https://ramenm.com/en/index.md
- Russian markdown: https://ramenm.com/ru/index.md
- Simplified Chinese markdown: https://ramenm.com/zh/index.md

## Machine-readable access
- HTML pages negotiate to markdown when requested with `Accept: text/markdown`.
- Every localized site page also has a URL-based markdown fallback at `/index.md` relative to the HTML page.
- llms.txt: https://ramenm.com/llms.txt
- llms-full.txt: https://ramenm.com/llms-full.txt