Mike Keeman

AI R&D | Safety | Interpretability

Engineer. Scientist. AI-safety advocate.
I architect systems where others see impossible constraints. Built production AI for government healthcare, impacted 20M+ people in Eastern Europe, earned PhD while running national R&D at 29. Bootstrapped and exited AI recruitment platform. Now diving deep into what makes AI truly intelligent.

The Journey

I started my career in sports sports - earned the PhD at the intersection of Data Science and Olympic sports - we tried to help athletes win more medals saving more health while trying.

I felt in love with AI, and jumped into healthech AI - founding a AI Research Lab in Medical Biophysics Centrem, workign in oen of the most 'sensitive' fields of healthcare - atomic medicine. Loads if NDA work, but simply saying - the harm that ARC reactor cuased to Tony Stark in Iron Man II - that was the area we'd had to study.

In 2022 I left Russia, restarred my career and dived into commercial application of AI. Hattl - incredible fast bullrun. Novel aproach to hiring. Bootsrap. Rapid Exit in late 2024. Something to be proud of.

But thats not the enf of journey. From 2025 I'm back to the frontier of AI. Intent driven orchestrators. Inteligent tools execution. Open source AI Personality research...

Nope, that's definitelly not the end That's just the beginning.

Selected Work

hattl - AI Recruitment Platform
Architected and shipped production AI recruitment platform from concept to acquisition.

Exit Achievement:
Bootstrapped to technology asset sale to US strategic acquirer
No VC backing - real customers, real revenue, acquired for proprietary tech value

AI & Systems Architecture:
Built proprietary matching engine using ML embeddings and cosine similarity matrices
Designed cascade of asynchronous ML tasks analyzing gamified assessments and job descriptions
Achieved sub-9 second fit calculations handling thousands of concurrent computations
Implemented domain-driven architecture (Django 5 + Next.js 14) with 99.99% uptime

Technical Leadership:
End-to-end system design from MVP to production-grade platform
Cloud-native AWS infrastructure (EC2, RDS, S3, CloudWatch) optimized for cost and scale
Mobile-first adaptive UX/UI with unified design system and reusable components
White-label middleware enabling customizable branded client portals
Outter - Intent-Driven Intelligence
AI orchestration platform that breaks conversational intelligence into specialized components, each optimized for specific tasks, then intelligently combines their outputs to generate superior responses. A few cents call to o1-mini with Outter consistently outperforms a 20x more expensive call to o3-pro without it.

Core Platform Development:
- Building proprietary AI orchestration engine - intent-driven components with intelligent output synthesis
- Developed custom intent analysis system using transformer architectures and specialized fine-tuning
- AI safety research - personality drift detection and mechanistic interpretability for production use
- Multi-modal execution framework - API + web automation + custom connectors with safety mechanisms
- Financial services integration - built complex billing engines with real-time usage metering and audit trails
AI Personality Drift Simulator
Open-source platform for AI personality drift study using mechanistic interpretability techniques. Optimized for local experiments on Mac M-family with local Llama-3.1-8B.

Core Innovation:
- First-of-its-kind mechanistic interpretability with psychological assessment for personality drift detection
- Clinically-validated assessment tools (PHQ-9, GAD-7, PSS-10) for AI behavior
- Real-time drift monitoring with statistical significance testing
- Automated intervention protocols based on neural circuit analysis

Research Capabilities:
- Neural circuit tracking and attention pattern analysis
- Clinical score interpretation with severity classifications
- Real-time drift monitoring with statistical significance testing

Background

  • PhD, Applied Data Science in Sports Medicine (2017)
  • Youngest Chief R&D Officer at National Research Centre at 29
  • Legacy Award for Sudden Cardiac Death in Sports ML-prediction algorithm, 2019
  • Global Talent UK - endorsed by Tech Nation, 2023

Tech Stack

Production AI Systems

Transformers (Hugging Face)
LLaMA
Mistral
Fine-tuning
LoRA/QLoRA
PyTorch
scikit-learn
Custom models
Embeddings
LangChain
Intent analysis
Action orchestration + rollback
Qdrant
FAISS
Semantic search
Knowledge management
RAG

Full-Stack & Infrastructure

Python 3.12+
FastAPI
SQLAlchemy 2
PostgreSQL
Async systems
React 19
Next.js 15
WebSockets
Azure
AWS
Docker
CI/CD
Monitoring
Zero-downtime
Perf tuning
Scalable arch