AI-Driven Engineering Leadership
I come into mature organizations to fix what's broken or build order out of chaos — and I build the systems that do it, out of people and agents, where AI absorbs the rote layer and people operate at judgment. Coding with AI is table stakes now; the scarce work is making the delivery and management layer AI-native, where outcomes are won or lost. I stay a hands-on engineer — I design and run these systems myself, not just describe them.
Languages: English (professional working), Russian (native)
Two AI systems in two different domains — a proven pattern, not a one-off. Coding with AI is table stakes; the scarce work is making the delivery and management layer AI-native, where outcomes are won or lost.
Built solo · running on live engagements
The system holds the full state of an engagement from the first pre-sales call: every call, email, document and commitment, captured across channels and reconciled. It weighs each fact by who is authorized to state it, keeps provenance on every claim, and flags contradictions, unkept promises and aging questions before they cost margin.
Public · self-demonstrating
An AI system that runs market intelligence and content operations end to end: reads signals, decides the next move, ships. Same engineering discipline, in the open — a live demonstration of the systems I build.
And beyond that — a multi-model video pipeline (character, scenes, TTS, whisper-timed keyframes, image-to-video, camera motion) and a production-grade agent/MCP toolchain — all of it built and running in production.
Evidence, metrics, and business outcomes.
$680M-revenue retailer · 20+ teams · 40+ services · 15M customers
Event-driven architecture, AI in production, and team enablement—with a multi-party integration layer so independently owned teams and services stay aligned on shared contracts; integration seams are explicit, not ad hoc.
1500+ employees · Enterprise wins · Predictable delivery restored
An engineering department in a B2B integrator under pressure: unstable delivery, strategic contract risk, and margin pressure.
Revenue +50% · 65k → 1M+ users · Weekly releases · 70+ test environments
Rebuilt the product lifecycle and engineering processes for a scalable transition to a cloud model.
199M+ profiles · ML in production · Cashflow + retention
Modernized the personalization platform with ML and predictive analytics to remove infrastructure bottlenecks.
Recommendations
From colleagues and managers on my LinkedIn profile.
"Grigoriy consistently demonstrated exceptional technical leadership and deep expertise as an Enterprise Architect. He was one of the key drivers behind the adoption of event-driven architecture and microservices across Askona's IT landscape — pragmatic, collaborative, and always focused on the right solution for both business and technology."
"I highly recommend him as an exceptionally competent professional in Enterprise Architecture, Solution Architecture, AI-driven innovation, and technical leadership. He led the company's strategic AI initiative, driving the adoption of AI tools and practices across the organization."
"He showed himself as an excellent technical leader. He put the development process to a new level, and the best practices he taught the team significantly improved the quality of the product."
"One of the best experts in software development I know. His expertise in application architecture, project management and team leadership is amazing. I'm really thankful to him for being my mentor."
"Open-minded and highly communicative. His influence on people and his professionalism made the development team one of the best I know."
"Grigoriy has excellent skills in the Linux command line, the automated deploy processes and distributed systems management."
Engineering breakdowns
01 · Facebook · 105k views
Book-as-context: architecture grounded on Tanenbaum
Ask for "ERP integration" → agent outputs FTP. The root isn't the model — it's missing domain context.
04 · GitHub · ★8 / 4 forks
Async MCP server with job queue
A naive MCP server blocks on long operations. Submit → poll, and 404-after-completion as a deliberate trade-off.
09 · YouTube · 3,600 views
Sending to Kafka is a two-phase commit
"Fire and forget" — wrong mental model. Dual write without outbox is a deferred incident.
07 · Governance
Cursor rules as governance
When 10 people vibe-code, standards drift. A guardrail in the toolchain, not a quarterly lecture.
I run a production MCP server that represents my professional profile. Connect it to your Claude and query my experience directly — a live demonstration of the systems I build.
https://mcp.dobryakov.net/mcp
"dobryakov-expert": {
"type": "http",
"url": "https://mcp.dobryakov.net/mcp"
}Different stakeholders get different, measurable value.
When you need an engineering organization where AI is the operating model — not one department's experiment — I take ownership of the whole function and free your time for strategy. An agent is a multiplier of your organization, not its competitor: without an owner who designs and owns the outcome of the whole system, agents turn into costs; with an owner, they turn into an operating model.
When you need one person you can trust with people, AI architecture, and delivery — not two separate roles — I cover all three levels with unified accountability for production outcomes. The more mature your agent stack, the more valuable the meta-role: who decides which agents run which tasks, where to stop them, and who owns the outcome of the whole system operating as one model.
When the AI roadmap exists but its translation into engineering practice stalls, I bridge the gap between architectural decisions and how teams actually work every day.
When the digitalization roadmap needs to land as outcomes, I align business goals with engineering execution without losing control.
A rare combination: engineering leadership + AI systems expertise in one profile — fills roles that would otherwise require two separate people with coordination overhead between them.
When architectural integrity must survive AI transformation, I embed new practices without breaking the foundation or creating unmanageable AI technical debt.
AI as an operating model — not one-off prompts, but a repeatable operating layer across three dimensions: SDLC, product, and operations.
When I take on an engineering organization, the first 90 days go into building the system — reading team & SDLC / product / operations, then wiring people and agents into one operating model with explicit owners, metrics, and stop criteria. Pilots reach production or get shut down with a reason. That's the cadence, whatever the scope.
A single hub with all public assets: blogs, videos, courses, community, project, and profiles.
Practical writing on project management, engineering leadership, software delivery, and applied AI in production contexts.
Public video materials on engineering leadership, architecture, and practical AI operations.
Where I post and where the professional record lives.
The clearest picture of how I think is in the cases, the engineering breakdowns, and Next Move Engine, my own AI system running the same patterns I ship for others. Open work is on GitHub.
Connect on LinkedIn