SKAI
An AI agent supporting corporate governance workflows with routing, tools, retrieval and answer verification.
Role: architecture and engineering
Open caseRUKazakhstan / Remote collaboration
Arman Aubakirov
Senior AI Systems EngineerI design and ship AI systems that connect models, data, backend infrastructure and a usable product.
From LLM inference, agent memory and voice interfaces to production deployment and measurable evaluation.
MODEL + MEMORY
+ RUNTIME + PRODUCT
LLM AGENTS VOICE DATA+ RUNTIME + PRODUCT
70B → 3-bitLLM optimization
NVIDIA H200inference infrastructure
1,144clinical protocols
Rust + Gosystems backend
Build / 04
Systems with a clear status and proof point
Each case separates public evidence, personal contribution, current limitations and the next experiment.
LLM Optimization
Inference and quantization work for large language models on NVIDIA H200 infrastructure.
70B class / H200
Open caseRUCortexDB
A Rust database for agent context, memory provenance and verifiable facts.
v0.2.0-beta.2
Open caseRUAI Therapist
A specialist-support system using clinical protocols, semantic retrieval and human review.
1,144 protocols
Open caseRUBuild / Research / Teach
One engineering method, three modes
Detailed case studies, articles and video lessons are currently available in Russian and are marked RU.
01 / Build
Production AI systems
Full product architecture from interface and API to data, models and operations.
02 / ResearchModels, memory and context
Reproducible experiments with explicit controls, limitations and verdicts.
03 / TeachPractical AI engineering
Eighteen open video lessons based on real implementation workflows.
Research
Current research questions
Measured prototype
Ternary Whisper
How far can speech-model weights be discretized while preserving measurable recognition quality?
Active research
Validated agent memory
How can an agent distinguish durable facts from stale or unsupported memory?
Research
Real-time digital humans
How can STT, LLM, TTS and lip-sync stay natural under a strict latency budget?
Teach / 18 videos
Practical learning paths
Detailed case studies, articles and video lessons are currently available in Russian and are marked RU.
Arman Aubakirov
Engineering beyond an API call
I work on problems where an AI model is only one part of the system: access control, context quality, runtime safety, observability and user recovery matter just as much.
Next step
Building a complex AI system?
We can start with the user outcome, constraints and one measurable vertical slice.
Discuss a project