Head of AI Engineering

Summary

Engineer and technical founder with 25+ years of systems experience, spanning low-level embedded/automotive software through to production AI systems leadership. Currently the primary architect of production AI engineering across a multi-product portfolio: designed and shipped a multi-model LLM pipeline (Squirrel Wisdom) achieving a 25x reduction in cost per generated course, and built Akinakes, a production multi-agent orchestration platform with full execution audit trails. Academic background in AI predates the current wave — postgraduate research on probabilistic-grammar text extrapolation, a precursor to modern language model architecture.

Flagship Projects

Akinakes — sole architect and engineer. Production multi-agent orchestration platform.

• Agent Bus: publish/match/route/deliver with capability-based routing; every message logged and traceable by chain ID through request, execution, and document change
• Human-in-the-loop governance: a Human Gate that parks execution chains pending approval, with five declarative autonomy levels from Manual-Only to Full-Auto-Sandbox
• Skill Registry of execution primitives with declared input/output schemas, permissions, autonomy levels, and audit requirements; versioned shared-memory document layer
• v0.8 — 641 passing automated tests across 10 development phases, 48 agents in registry

Squirrel Wisdom — founder, sole AI engineer, architecture through product direction. AI-native education platform: teachers describe a course in plain language; the system generates a conversational chatbot-tutor, a structured textbook, and TTS audio from a single authoring pass.

• Self-correcting, multi-stage multi-model LLM pipeline — lightweight models handle generation, validation, cleaning, and repair; large models (Claude Sonnet, GPT-4/5) reserved for narrative quality — a 25x reduction in cost per generated course
• Eval-first pipelines with regression tracking and automated failure categorization; caught multiple silent model degradations across Claude, GPT, and LLaMA versions
• RAG + curriculum graph for course-aware conversational agents with retrieval-quality tuning and hallucination mitigation
• Integrated and benchmarked 5 TTS providers (Azure, Edge TTS, Piper, OpenVoice, Silero) with content/context-based routing; deployed to 3 pilot schools, ~50 courses, ~100 active users

ImmersiveLion — founder / engineering lead (Task Force 38). Virtual-school platform built on Unreal Engine 5 with AI-driven NPC characters, up to 100 concurrent users.

• Built the server-side AI backend: real-time NPC behavior engine, Claude-powered dialogue management, OpenAI TTS speech pipeline — became the architectural foundation of Squirrel Wisdom's production LLM pipeline
• Fault-tolerant AI workload infrastructure with observability instrumentation, feature drift detection, and rollback paths, deployed on self-hosted bare-metal

Camerlengo — architect. Distributed backend providing authentication, session management, messaging, storage abstraction, LLM-adapter routing, and auditability as the common foundation under the product portfolio.

AI & LLM Expertise

• Multi-model orchestration (Claude Sonnet, GPT-4/5, LLaMA via OpenRouter)
• RAG, vector search, and retrieval systems
• Structured outputs, prompt and context engineering
• Self-correcting / self-repairing multi-stage LLM pipelines
• Eval-first development, regression tracking, automated failure detection
• Human-in-the-loop governance, AI safety controls, and full execution audit trails
• Multi-agent systems and agentic workflow design
• Speech AI: TTS/ASR pipeline design and provider benchmarking

Relevant Experience

IT Partners Solutions / NavLink.Net — Founder / Head of AI Engineering (2022–present). Primary technical driver and de facto head of AI engineering across an AI product portfolio (Akinakes, BeautySqrl, Squirrel Wisdom, Camerlengo): architecture, implementation, deployment, observability, and governance. Designed the multi-model LLM routing architecture behind the 25x cost reduction. Direct management of a 5-person engineering team.

Task Force 38 — Founder / Engineering Lead (2023–present). Embedded electronics company; led ImmersiveLion and its AI backend; built AI-driven automation systems integrating real-world data streams with LLM-based processing. Direct management of a 6–8 person engineering team.

BioIntelliSense Inc — Senior AOSP Engineer / Product Owner (2020–2025). Product owner for an FDA-regulated medical wearable; owned the complete Android/AOSP system stack and backend workflows for continuous biosignal data streams — the same engineering discipline extended to safety-critical, audited systems.

Why This Is Different

Most AI engineers started with the model API. This background started below the operating system — real-time embedded firmware, AOSP framework internals, automotive safety systems, FDA-audited medical devices — and grew into AI systems leadership from there. That order matters: it means production AI systems that are engineered, instrumented, and governed like the safety-critical systems they increasingly are, not prototypes wrapped around a demo.

Education & Recognition

• M.Sc. Computer Science, Saint Petersburg Electrotechnical University (LETI) — Diploma with Honors, GPA 4.96/5.0
• Postgraduate studies (PhD track), Artificial Intelligence, LETI — probabilistic-grammar text extrapolation, an early precursor to modern language model foundations
• 5 granted patents (software systems and embedded technologies)
• Author, The World Inside Out: How to Be a Programmer Without Writing Code

Contact

Open to Head of AI Engineering, Principal/Staff AI Engineer, and AI R&D leadership roles.
Email: konstantin@khait.org · Phone: (919) 805-4708 · Location: USA / Europe