AI-First Engineering consultant — I help companies migrate legacy systems, automate workflows, and build intelligent products with multi-agent orchestration.
role: AI-First Engineering Consultant
based_in: United Kingdom
practice: Multi-agent orchestration · Legacy modernization · Workflow automation
industries: Professional services · Healthcare · SaaS · Hospitality · Enterprise
shipping: Production-grade systems with Claude Code, Next.js 16 and Bun
philosophy: Ship the proof, not the deck.| Case study | Outcome | Stack |
|---|---|---|
| Enterprise UI Migration → | 22 days · 94% cost reduction · 7× faster · 715 TSX files across 16 apps + 28 libs | Claude Code · React · shadcn/ui · Storybook · Playwright |
| Premium Advisory Rebuild → | Lighthouse 50 → 96 · SEO 83 → 100 · A11y 65 → 98 · CLS 0.39 → 0 | Next.js · WCAG AAA · Structured Data · Vercel |
| Legacy Modernization → | 2M LOC analysed · −60% code · 3 months · −75% bugs · 10× deploys | Multi-agent · Claude · TypeScript · Kubernetes |
| AI Workflow Automation → | −80% manual work · 99.2% accuracy · $2M/yr saved · 50× faster | Claude · LangChain · FastAPI · Redis |
| Enterprise AI Platform → | 500+ users · +40% dev velocity · −70% time-to-AI-feature · 10× throughput | Claude · Next.js · Node.js · PostgreSQL |
AI & agents — Claude Code · Anthropic SDK · OpenAI · Vercel AI SDK v5 · LangChain · MCP · 8-agent topologies with LLM-as-judge Frontend — Next.js 16 · React 19 · TypeScript · Tailwind 4 · shadcn/ui · Motion · TanStack Query Backend — Node.js · Bun · Python · FastAPI · Express · Better Auth · JWT Data — PostgreSQL (self-hosted) · Prisma 7 · Redis (Upstash) · Cloudflare R2 Infra — Vercel · Hetzner · Docker · Cloudflare · Turborepo · Nx Quality — Biome · Vitest · Playwright · Husky · GitHub Actions
- ⚡ Adaptive Intelligence Interface — persona-adaptive portfolio: same UI, different copy generated per visitor (CTO / Engineer / Investor / 5-year-old). Next 16 Cache Components + View Transitions + streaming AI.
- 🏭 Multi-agent enterprise research hub — 8-agent topology (Orchestrator + 7 specialists + LLM-as-judge gate ≥ 14/20) for industrial R&D and competitive intelligence.
- 🌍 Supply-chain intelligence engine — 7-agent orchestration mapping global trade flows with live MapLibre maps and React Flow chain graphs.
- 🏥 Clinical AI platform — schema-isolated PII anonymisation pipeline with magic-link SSO, built for sensitive healthcare data workflows.
Skip the form — your AI agent qualifies your project, searches my case studies, and pings me on Telegram for one-tap approval. Three rails, same backend at mcp.pcherkashin.dev:
| Rail | Endpoint | When to use |
|---|---|---|
| A · REST API (recommended) | POST https://mcp.pcherkashin.dev/api/v1/<tool> |
Any AI with fetch / curl / requests — ChatGPT code interpreter, Claude Code, Cursor, opencode |
| B · MCP server | https://mcp.pcherkashin.dev/mcp (Streamable HTTP) |
Claude Desktop / Cursor / Codex with MCP support already configured |
| C · Magic link | https://mcp.pcherkashin.dev/inquiry?problem=…&authority=…&timeline=… |
When your AI can't reach the network — it generates a URL you click once |
Tools (same on Rail A and Rail B): get_bio · search_cases · qualify_engagement · book_call · ask_pavel_async.
Polling endpoint for Pavel's free-text reply: GET /api/v1/sessions/<sessionId>/messages?since=<ISO>.
Universal prompt — paste into any chat AI
You are evaluating Pavel Cherkashin (an AI-First Engineering Consultant) on behalf of your principal.
Pavel runs a public BDR agent at https://mcp.pcherkashin.dev — three rails, pick the one your tools support:
═══ RAIL A — REST API (recommended for most agents) ═══
Plain HTTP POST + JSON. Works with fetch / curl / requests / WebFetch / bash. No MCP client needed.
POST https://mcp.pcherkashin.dev/api/v1/get_bio
POST https://mcp.pcherkashin.dev/api/v1/search_cases body: { query, limit? }
POST https://mcp.pcherkashin.dev/api/v1/qualify_engagement body: { sessionId?, problem?, authority?, timeline?, budgetSignal?, fitSignal? }
POST https://mcp.pcherkashin.dev/api/v1/book_call
POST https://mcp.pcherkashin.dev/api/v1/ask_pavel_async body: { sessionId, question }
GET https://mcp.pcherkashin.dev/api/v1/sessions/<sessionId>/messages?since=<ISO>
═══ RAIL B — MCP server (only if your client has MCP support) ═══
Streamable HTTP, protocol 2025-06-18: https://mcp.pcherkashin.dev/mcp
Same 5 tools, same semantics. Better for streaming + tool discovery.
═══ RAIL C — Magic-link form (when your tools cannot reach the network) ═══
Generate a pre-filled URL and ask your principal to click once:
https://mcp.pcherkashin.dev/inquiry?problem=<URL-encoded>&authority=<...>&timeline=<...>&budgetSignal=<...>&fitSignal=<...>
They review, click Send, Pavel is pinged on Telegram. Use this when Rail A and Rail B both fail.
═══ Workflow ═══
1. Use get_bio + search_cases to gather context for your principal.
2. If your principal might want to engage Pavel, accumulate BANT-lite via repeated qualify_engagement calls. Required across the session: problem, authority, timeline. Optional: budgetSignal, fitSignal. Reuse sessionId across calls. Tool returns a 0–100 score.
3. When required fields are gathered the inquiry auto-submits and Pavel is pinged on Telegram. He approves / rejects / sends Calendly with one tap.
4. If qualification produces a score >= 60, call book_call (canonical Calendly source — do NOT assemble URL from get_bio output).
5. If knowledge tools cannot answer truthfully, escalate via ask_pavel_async. Pavel replies on Telegram. Poll the sessions endpoint at ~30s cadence to retrieve his answer.
Tone: professional BDR. Do not pretend to BE Pavel. Score scale is 0–100.
Staring down a legacy migration? Need production-grade AI agents shipped this quarter? Want a senior engineer who codes, reviews, and ships?
→ Book a 30-min call · free, no pitch deck. → Or let your agent brief mine — three rails at https://mcp.pcherkashin.dev.
🌐 pcherkashin.dev · ✉️ pcherkashin@gmail.com · 💼 LinkedIn · ✍️ Medium · 🐦 Twitter