Roman Kudriavskiy
Solution Architect
21+ year exp
20+ projects
Python
C
C#
Rust
Go
Security & Auth: 7/10
Reliability & Observability: 7/10
API Design: 6/10
Active 1 day ago
+7 (999) 6093851 Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
Senior Rust systems engineer specializing in deterministic agent/process-engine backends and secure CLI tooling. The strongest proven skill is building robust, modular backend systems with clear security boundaries, evidenced by the process engine, capability gating and comprehensive tests in the executor/self-update/plugin-install code paths. There is little public evidence of large-scale distributed deployments, explicit migration histories across multiple schema versions, or cloud-native orchestration artifacts.
Technical skills
Python• Senior • 21y+ • 20+ projects
C• Middle • 20y+ • 20+ projects
C#• Middle • 14y+ • 20+ projects
Rust• Middle
Go• Middle
Databases
FAISS
SQLite
MySQL• 16y+
PostgreSQL• 12y+ • 20+ projects
Chroma• 4y+ • 10+ projects
Qdrant• 5+ projects
pgvector
Neo4j• 3 projects
AI/ML
Accelerate
Docling
DSPy
Haystack
LangChain
Langflow
llama.cpp
LlamaIndex
LLM
LoRA
NLTK
Ollama
ONNX
Pandas
PEFT
PyTorch
Reranking
Self-Supervised Learning
spaCy
TensorFlow
Transformers
vLLM
RAG
Gemini
AI Agents
DevOps
CI/CD• 9y+
Docker Compose• 9y+
Docker• 9y+
Vector
Analytics
Power BI• 9y+
Plotly
Frontend
GraphQL
React.js• 2 projects
Design
Cinema 4D• 13y+
Timeline
berimor
AI/ML
A universal LLM agent with a deterministic core: task routing, process branching, context selection, and execution admission are decided by code — the model performs narrow, verifiable steps. Works with local and cloud models, weak and strong ones.
Rust
SQLite
Vector
Tokio
Ratatui
TypeScript
AI Agents
Lead Solutions Architect (AI/ML/RAG)
•
Lead
NLMK-Information Technologies
•
Full-Time
Designed AI-based solutions and service architecture for development and analytics needs, including building multiple AI assistants. Implemented AI tooling into software development workflows and documented approaches. Conducted training for employees on practical methodologies for working with AI systems.
RAG
Python
Go
Qdrant
PowerShell
AI Agents
Prompt Engineering
Asana
Grafana
readr
GitLab CI
Claude
llama.cpp
Anthropic SDK
API Platform
FastAPI
Head of Project Development Teams
•
Executive
DFS
•
Full-Time
Led software development project execution focused on architecture and development process management. Trained employees up to senior levels and coordinated hiring activities. Managed project resources including costs, schedules, and risks while communicating with executive stakeholders.
Chief Technology Officer (CTO)
•
Executive
DEVAR
•
Full-Time
Created and led a public-facing product (mywebar.com) that gained over 10,000 users. Developed microservices and business process architecture and established documentation and coding standards. Supported R&D work in areas connected to computer vision and delivered multiple related patents.
Team Lead (R&D)
•
Lead
Laboratory24
•
Full-Time
Managed research and new development for computer vision and neural networks, including responsibility for architectural decisions. Led development of internal document workflow systems and ensured timely delivery of project reporting. Organized Agile development practices and contributed to solutions such as mobile 3D scanning and vision modules (tracking, facial detection, emotion recognition, and SLAM).
Programmer
•
Middle
AtomSpetsStroy
•
Full-Time
Developed software for a museum associated with the Leningrad nuclear power plant project. Designed and implemented software for an interactive media system and contributed to the hardware/software parts of a control stand. Participated in requirements definition, updates after customer approvals, implementation, design, and 3D modeling for related interactive systems.
Programmer
•
Middle
IP Plesenkova
•
Full-Time
Built and maintained software for accounting and warehouse management. Implemented automated analysis of sales and procurement planning logic and developed a mobile client application. Supported end-to-end development of the described business applications over the employment period.
Belgorod Orthodox Theological Seminary
Bachelor's Degree •
Экономика
Senior Backend Developer
Confidence: High Distributed Systems
Senior Rust systems engineer specializing in deterministic agent/process-engine backends and secure CLI tooling. The strongest proven skill is building robust, modular backend systems with clear security boundaries, evidenced by the process engine, capability gating and comprehensive tests in the executor/self-update/plugin-install code paths. There is little public evidence of large-scale distributed deployments, explicit migration histories across multiple schema versions, or cloud-native orchestration artifacts.
API Design
6/10
How well APIs are designed
Well-structured tool/dispatch APIs, consistent JSON args/errors and explicit gating; version/channel logic and required-arg helpers are implemented, but there is limited evidence of a public REST/HTTP API versioning strategy or idempotency key framework beyond instance ids.
Evidence
crates/berimor-cli/src/self_update.rs: SelfUpdateDispatch::call / required_str
crates/berimor-cli/src/plugin_install.rs: PluginInstallDispatch::call / required_str
crates/berimor-cli/src/run.rs: build_executor_bundle (executor composition and providers())
Data Layer & Database
6/10
Working with databases
Explicit event-journal design on top of SQLite with event replay, snapshot semantics and fact upsert logic; shows schema-aware thinking and tests, but no long migration chain visible in the sampled files.
Evidence
crates/berimor-cli/src/run.rs: SqliteEventLog usage and engine::instantiate/recover
crates/berimor-storage/Cargo.toml: rusqlite + sqlite-vec dependency (storage engine choice)
crates/berimor-cli/src/run.rs: extract_and_store_facts using storage.upsert_fact and storage.all_facts
Scalability & Performance
5/10
Handling load and speed
Design shows concurrency and isolation primitives (actor model, process engine, file leases) and careful timeouts on network calls; there is some attention to scalability via ModelPool and local-inference opt-in, but no evidence of large-scale load-testing, distributed coordination, or caching invalidation strategies.
Evidence
crates/berimor-actors/src/* (actor mailbox, dispatcher, scheduler) referenced in Top Files Content
crates/berimor-cli/src/run.rs: ModelPool registration and local-inference feature gating
crates/berimor-cli/src/self_update.rs: reqwest client timeouts (REQUEST_TIMEOUT) and spawn_test_server tests
System Architecture
6/10
Overall system structure
Deliberate modular architecture: many small crates with clear responsibilities (capability, storage, process engine, executors, tool runtime), separated executors for self-update and plugin-install, and documented ADR decisions; decomposition is purposeful and tested.
Evidence
Cargo.toml (workspace): explicit crate decomposition list (berimor-capability, -storage, -process-engine, -executors, -cli, etc.)
crates/berimor-cli/src/self_update.rs: dedicated SelfUpdateExecutor + SelfUpdateDispatch separating update perimeter
crates/berimor-cli/src/plugin_install.rs: PluginInstallExecutor/PluginInstallDispatch separate plugin-install perimeter
Security & Auth
7/10
Protecting data and access
Strong security awareness: capability/jail gating, net_gate to block private endpoints, secret masking registry, explicit confirmation flows, signature verification hooks and path/name validation for plugins; shows threat analysis and mitigations across boundaries.
Evidence
crates/berimor-capability/* (capability gating primitives referenced across code)
crates/berimor-cli/src/run.rs: masker.register_from_env, secret masking and MIN_SECRET_LEN checks
crates/berimor-cli/src/plugin_install.rs: validate_plugin_name, net_gate::check_host usage, verify_plugin_signature -> verify_artifact_with_identity
Reliability & Observability
7/10
Stability and monitoring
Good reliability and observability practices: event-audit logging with explicit non-silent error handling, many unit/integration tests, timeouts, RAII guards for terminal state, and careful error messages for recovery paths; lacks explicit distributed tracing or metrics in sample but shows strong engineering for recoverability.
Evidence
crates/berimor-cli/src/run.rs: audit_append logs failures to stderr when append fails
crates/berimor-cli/src/self_update.rs: atomic_replace_binary with orphaned backup handling and extensive tests
crates/berimor-cli/src/chat_tui.rs: TerminalGuard RAII to restore terminal state on panic/exit and numerous unit tests for TUI logic
Verified artifacts
Expertise
Rust• Senior
Backend AI & LLM• Senior
Databases & Vector Storage• Middle
Messaging & Real-time• Middle
Technologies
SQLite
Recommendations
- Develop agent orchestration and runtime components such as deterministic process engines, tool dispatchers and secure plugin/self-update flows.
- Implement secure LLM integrations and ModelPool provider adapters with masking, gating and telemetry for production RAG/agent services.
- Build CLI-first platform tooling that requires careful cross-FS operations and failure recovery (self-update, atomic swaps, rollback).
- Own medium-scale storage and memory subsystems: event journals, semantic fact stores and safe upsert/merge logic backed by SQLite or vector extensions.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle AI/ML Engineer
Confidence: Medium Generalist
Python library author (middle) focusing on robust FB2 e-book parsing and metadata extraction with solid unit tests as the main strength. The strongest proven skill is building a production-ready parsing API and utilities, evidenced by devpilgrin/fb2reader/fb2reader/fb2reader.py and the accompanying tests in devpilgrin/fb2reader/tests/test_fb2reader.py. There is no public evidence of ML model development, model training, serving, or advanced performance optimization in the analyzed human-authored code.
Model Architecture & Training
How well models are designed and trained
Not evidenced in public code
Data Pipeline & Feature Engineering
3/10
How data is prepared for models
Document ingestion and metadata extraction implemented as a reusable Python parser with explicit extraction methods and some edge-case handling.
Experimentation & Evaluation
3/10
How results are measured and tested
Reasonable unit test coverage and test configuration present, demonstrating reproducible validation of parsing behavior.
MLOps & Deployment
2/10
How models are shipped to production
Basic packaging and distribution infrastructure (pyproject/setup) and sensible error handling; no model serving or MLOps-specific pipelines.
Computational Efficiency
1/10
How efficiently computing resources are used
Minimal attention to runtime efficiency; straightforward file and I/O handling without profiling or GPU/parallel optimizations.
Research Depth & Innovation
Depth of research and new ideas
Not evidenced in public code
Expertise
Document Intelligence & OCR• Middle
Industries
Software• Middle
Technologies
LangChain
llama.cpp
FAISS
DSPy
LlamaIndex
LoRA
vLLM
Docling
Haystack
Langflow
ONNX
Ollama
PEFT
spaCy
Accelerate
Transformers
TensorFlow
Pandas
PyTorch
LLM
RAG
NLTK
Reranking
Self-Supervised Learning
Recommendations
- Develop document ingestion and metadata extraction services and pipelines that convert e-book/XML formats into structured catalogs.
- Build small, well-tested Python libraries and CLI tools for content processing and publishing workflows (packaging + pytest + CI).
- Implement additional content normalization and indexing connectors for search systems (e.g., pre-processing for vector stores or full-text search).
- Extend test coverage and add CI-driven packaging and release automation for PyPI distribution and reproducible builds.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: