Andrey Reshetnikov

Roles I am open to

Product-minded AI roles that combine discovery, hands-on prototyping, evaluation, RAG and agent workflows, engineering handoff, and adoption inside business teams. Remote, or hybrid by discussion.

• Forward Deployed Engineer
• Forward Deployed AI Strategist
• Applied AI Engineer
• AI Engineer
• Prompt Engineer
• AI Prototyping Manager / Lead
• AI Product Manager
• AI Solutions Engineer / Architect
• Agentic Workflow / RAG Systems Lead
• LLM Evaluation Lead
• GenAI Implementation Manager
• AI Transformation Manager / Lead
• AI Enablement Lead
• AI Outcomes / AI Success Lead
• AI Strategy Lead

Contact

Use LinkedIn or direct Telegram.

Prompt / AI Engineer, Tochka Bank

January 2025 to present.

• Built a Hermes-based prototyping lab that turns raw materials, evaluation rubrics, and stakeholder notes into live Telegram pilots, reusable scenario packs, role-play and evaluator prompts, privacy rules, and Git-versioned artifacts. Users test the LLM experience before dedicated UI or backend investment; only validated concepts move into development, backed by interaction traces and clearer engineering requirements.
• Built pipelines for preparing tax-law materials for AI Accountant RAG, reducing context volume by about 50 percent without loss of recall and improving retrieval and inference efficiency.
• Created an LLM-as-judge evaluation pipeline against expert-labelled golden datasets, enabling fast regression checks for prompts and pipeline checkpoints without repeating full human evaluation cycles.
• Developed AI Seller analytics and a LangGraph-based multi-agent assistant for marketplace sellers with about 100 financial metrics, RAG, web search, answer validation, typed Pydantic contracts, and automated PDF reporting.
• Introduced an AI-native operating model for business teams: recurring task to instruction or template to reusable Codex or Claude Code skill to local runtime to service or MVP. Ran practical Codex meetups for product managers and training for business units on Codex, Claude Code, GitLab, and context engineering.
• Implemented scenario-based evaluation and a controlled improvement loop: sessions and traces to curator analysis, change proposals, draft versions, human review and promotion, and sanitized Git export.
• Designed multi-agent workflows with Hermes and OpenClaw and custom MCP integrations for ChatGPT and Claude, covering intake, routing, approvals, tracing, retries, receipts, and fail-closed quality gates.

Gen AI apps implementation manager, NDA web3 / Crypto / GameFi company

May 2023 to November 2024.

Gemba and Value Stream Mapping first, GenAI tools second.

• Taught coworkers to use LLMs on company data, reducing outsourced copywriting needs by 80 percent.
• Reported that the production center shot about five times less video, marketing A/B tests started about four times faster, and design teams produced content about three times faster with diffusion models including Stable Diffusion and Kohya_SS.
• Designed a support and sales system using HelpDesk, chatbots, make.com, and low-cost translation: five-second first response around the clock, instant user engagement and onboarding, and 50 percent of sales attributed to the bot.
• Turned customer queries into topic modelling and product-feature estimation to improve backlog value.
• Built an internal BLIP-2 text-based image-search tool for faster visual-content retrieval and less content-creation redundancy.

Management consulting, venture, and technology startup roles

2016 to 2023: Strategy Partners Group / Skolkovo Foundation / Philip Morris International / Level.Travel / fuel4growth.vc / test-gear.io.

Combined business-process redesign, product and venture development, customer discovery, market validation, financial modelling, and strategic analysis.

Public work: New Runtime

Agent-readable New Runtime public surface: Designed and shipped a read-only public agent surface for newruntime.com: llms.txt, full agent guide, OpenAPI route contract, read-only MCP resource manifest, JSON/Markdown records, stable IDs, provenance fields, source URLs, timestamps, and mutation probes.

Evidence-linked AI newsroom and systems atlas: Built New Runtime as a static, crawlable AI newsroom and knowledge atlas with posts, raw signals, patterns, topic hubs, projects, systems, source observatory, public source ledger, production retrieval checks, and agent-index contract tests.

Operational agent boundary design: Documented and tested owner-gated operational boundaries for Hermes/OpenClaw-style workflows: read-only public routes, no public write actions, no deploy or publication tools, explicit approval requirements, traces, rollback, and sanitized public dossiers.

Core expertise

• AI/LLM systems: agentic workflows, multi-agent orchestration, RAG, MCP, structured outputs, context engineering, golden sets, LLM-as-judge evaluation.
• Product and management: discovery, hypothesis framing, PoC design, impact metrics, process redesign, stakeholder management, workshops, engineering handoffs.
• Engineering: Python, LangGraph, OpenAI APIs, Pydantic, Pandas, OpenPyXL, Git/GitLab, tracing, observability.
• AI-assisted development: Codex, Claude Code, Cursor, reusable skills, versioned prompts, review gates, auditable artifacts.
• AI research: Author of t.me/qwgai; continuous analysis of LLM tooling, agent architectures, MCP, evaluation practices, and AI product developments.

Education

• Moscow Institute of Physics and Technology: Tech Entrepreneurship.
• Moscow Power Engineering Institute: Power Grids.