About Me

Production AI systems for companies that hate slideware. From SMB ops to Fortune 500 engineering.

20+
Years Engineering Software
30+
AI Projects Shipped
50+
Self-Hosted Production Tools

AI-First Engineering Consultant

Based in Derry~Londonderry, working with companies across Northern Ireland, Ireland, the UK, and remotely worldwide — from solo founders to Fortune 500 engineering orgs. My focus: find the highest-ROI AI levers in your operation and ship systems that survive production, not demos that survive a meeting.

Two decades shipping software. 30+ AI projects since 2022. End-to-end delivery — discovery, architecture, multi-agent implementation, governance. Copilots, workflow automations, voice agents, research engines, legacy modernization at 2M+ LOC scale — I've shipped them all.

Recent Results

All cases →
94% cost reduction · 7× faster

Enterprise UI Migration

715 TSX files migrated to shadcn/ui across 16 apps and 28 shared libs in 22 days. Claude Code did 90% of the work; humans focused on PR review and architecture checkpoints.

Read case →
2M LOC analysed · −60% code

Legacy System Modernization

Multi-agent codebase analysis, dependency mapping, and automated migration plans. Shipped in 3 months. 75% fewer bugs, 10× deploy frequency post-migration.

Read case →
$2M/yr saved · 99.2% accuracy

AI Workflow Automation

Multi-agent document processing pipeline for healthcare. Replaced 80% of manual workflow at near-perfect precision. 50× faster than the prior system.

Read case →

What I Build

  • Multi-agent systems — 8-agent topologies with Claude Code, LLM-as-judge quality gates, and production budget controls. Up to 7× faster delivery on real migrations.
  • Legacy modernization — Multi-agent codebase analysis, dependency mapping, and automated migration plans for 2M+ LOC systems. 60% code reduction is typical, not aspirational.
  • AI products & copilots — Chat, voice, and web agents that book demos, answer from your docs with citations, and update your CRM/spreadsheets in real time.
  • Workflow automation — Email triage, document drafting, CRM ops, weekly KPIs. 80% manual work eliminated at 99.2% accuracy on real healthcare pipelines.

Why Work With Me

Production-grade, not demos

Multi-agent budget controls, schema-isolated PII, audit trails, abort handlers, retry-with-fallback on every LLM call. Real systems with real guardrails — not notebook prototypes that die under load.

End-to-end delivery

Discovery → architecture → multi-agent implementation → governance. I don't hand off slides; I ship code that survives your next deploy and the one after that.

8-agent orchestration topologies

Production multi-agent systems with LLM-as-judge quality gates (≥14/20 rejection), Claude Code orchestration, and per-call budget enforcement. Not "an LLM call wrapped in a Lambda".

Two decades of shipping

Senior engineering judgment, not AI hobbyism. I know what production failure looks like at 3am — and how to architect so it doesn't happen.

How Engagements Work

1

Discover

2-hour session to map quick wins and the highest-ROI AI levers in your operation. Walk away with a 90-day plan you can execute solo or with me.

2

Pilot Sprint

Build and demo a working prototype in a real workflow within 1–2 weeks. If it doesn't move the needle, we kill it — no sunk-cost theatre.

3

Implement

Scale what works, train your team, ship governance policies. End-state: your team owns it, with or without me on the call.

Industries I Serve

Professional Services
Legal · Accountancy · Tax Advisory
Healthcare
Clinical workflows · PII-grade pipelines
Software & SaaS
Product teams · Engineering orgs
Hospitality & Tourism
Hotels · B&Bs · Attractions

Brief Pavel via your AI

New

Skip the form. Three rails — pick the one your AI tools support. Each one qualifies your project, searches Pavel's case studies, and pings Pavel on Telegram for one-tap approval.

RAIL A

REST API

Plain HTTP — works with any agent

https://mcp.pcherkashin.dev/api/v1

POST /qualify_engagement etc. Same 5 tools, JSON in/out. Works in ChatGPT code interpreter, Cursor, Claude Code bash, opencode, anywhere with fetch.

RAIL B

MCP server

Streamable HTTP, JWS-signed agent card

https://mcp.pcherkashin.dev/mcp

Protocol 2025-06-18. Use this if your client (Claude Desktop, Cursor, etc.) has MCP support pre-configured. Better tool discovery + streaming.

RAIL C

Magic link

Zero AI-side networking — just a URL

https://mcp.pcherkashin.dev/inquiry?problem=...&authority=...&timeline=...

When your AI can't reach the network, ask it to generate a pre-filled URL. Click once, review, send. Pavel pinged on Telegram.

Universal prompt — paste into any 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 http…

2314 chars · covers all three rails. Drop it into ChatGPT / Claude / Cursor / Gemini and the AI picks the right path automatically.

How it works. Your agent runs the qualification, Pavel reviews on Telegram with one-tap Approve / Reject / Send Calendly. No email tag. No form. No middleware between you and decision-grade context.

Tech Stack

ai
Claude CodeAnthropic SDKOpenAIVercel AI SDK v5LangChainMCP
frontend
Next.js 16React 19TypeScriptTailwind 4shadcn/uiMotion
backend
Node.jsBunPythonFastAPIBetter Auth
data
PostgreSQLPrisma 7RedisCloudflare R2
infra
VercelHetznerDockerCloudflareTurborepo

Staring down a legacy migration? Need production AI agents shipped this quarter?

Book a 30-minute call — free, no pitch deck.

Book a 30-min CallConnect on LinkedIn