Kantemir Satibalov
AI Engineer
4+ years ML exp
5+ projects
Python
Go
SQL
Active 23 days ago
+7 (918) 0045244
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Career analytics
Roles
Overview
AI engineer (senior-level) focused on document-grounded retrieval and API orchestration for RAG systems. The strongest proven skill is building and integrating a production-minded retrieval stack and service boundary (Python RAG HTTP/gRPC service + Go orchestration) as shown by api/http/app.py and grounded-horticulture_ru/server/rag_chat.go. There is limited public evidence of large-scale distributed systems work like sharding/consensus, measured performance tuning at scale, or an exhaustive cloud IaC/ops pipeline in the analyzed human-authored files.
Technical skills
Python• Middle • 3y+
Go• Middle
SQL
Python
Gunicorn
FastAPI• 3y+
Databases
Redis
pgvector• 3y+
PostgreSQL• 3y+
Qdrant• 3y+
Chroma
AI/ML
RAG• 4y+ • 5+ projects
Embeddings• 3y+
Hybrid Search• 3y+
LangChain• 3y+
LLM• 3y+
Prompt Engineering• 3y+
AI Agents
Cursor• 5+ projects
NLP
Multimodal AI
DevOps
Rest API
IaC
Containers
OpenTelemetry
CI/CD• 3y+
Docker
Terraform
Prometheus
Kubernetes
Helm
Timeline
AI Platform Engineer
•
Middle
OOO Galereya-Maykop
•
Full-Time
Partially confirmed
30%
Architected and shipped Grounded LLM v0.3.0, a production RAG platform (Go + Python) serving document-grounded AI assistants with source citations and numeric verification (±0.01).
Achieved 100% fact-retrieval accuracy on a 99-case adversarial evaluation suite running in GitHub Actions CI.
Built grounded-vllm, an OpenAI-compatible proxy placing token-level guardrails in the LLM serving path with +0.71ms p50 overhead.
Developed grounded-guardrails (Rust + Go gRPC) — zero-copy token buffer, PII detection (<50μs), and numeric verification.
Created grounded-bench, a public benchmark with 1000 cases across finance, medical, legal, technical, and HR domains (NVR, CP, HR, RR metrics).
Designed grounded-agent, a ReAct orchestrator with multi-step reasoning, tool use via MCP Gateway, and Redis session memory.
Built mcp-gateway (Go) — MCP protocol gateway with WASM sandboxed tools (wazero), OpenTelemetry tracing, and Prometheus metrics.
Designed a multi-tenant REST API (OpenAPI/SSE) with tenant data isolation, authentication, and audit logging.
Integrated local LLM inference via vLLM (GPU, p95 ~0.7s) and Ollama (CPU, p95 ~4.5s) with Docker Compose profiles.
Implemented Redis caching for embeddings (1h TTL) and LLM responses (24h TTL) with X-Cache: HIT/MISS headers.
Delivered infrastructure-as-code via Docker Compose, Helm/Kubernetes, and Terraform (AWS/GCP/Azure).
Set up CI/CD pipelines with automated testing, security scanning (Trivy), and coverage gates (≥60%).
Optimized PostgreSQL schemas and queries, reducing analytics report generation time by 60%.
Built internal AI assistants for ticket classification, reducing L1 support workload by 40%.Designed an open-source LLM platform for grounded document assistants with cited and verified outputs. Built orchestration and a multi-tenant REST API in Go and Python/Flask, including authentication/authorization and streaming. Implemented hybrid retrieval (BM25 + embeddings with RRF) using Chroma, Qdrant, and pgvector, plus eval-driven CI quality gates. Delivered on-prem deployment using Docker Compose, Kubernetes, and Terraform across AWS/GCP/Azure.
Python
Go
PostgreSQL
Docker Compose
Chroma
Qdrant
pgvector
Hybrid Search
Embeddings
RAG
GitHub Actions
CI/CD
Kubernetes
Docker
Terraform
vLLM
huggingface_hub
MCP Protocol Gateway with WASM Sandbox
Go
gRPC
Jaeger
OpenTelemetry
Grafana
Prometheus
Kubernetes
Helm
Docker Compose
grounded-vllm
AI/ML
LLM Serving-Path Adapter with Guardrails Integration
Python
FastAPI
vLLM
Ollama
gRPC
Pydantic
Docker Compose
grounded-llm
AI/ML
RAG Platform with Hybrid Retrieval & Local LLM Inference
Go
Python
RAG
Flask
Helm
Qdrant
Trivy
Terraform
Redis
PostgreSQL
Docker
Career analytics
Verified work history and code activity in one view - open-source impact, work rhythm, trust signals and a forecast of when this candidate is likely to move.
1
Years active
215
Commits analyzed
Work rhythm
UTC+4
12 AM6 AM12 PM6 PM12 AM
Code activity
Low
Peak
Last activity
Aug 2026 · 2 Days
Active days (90d)
23
Typical work hours
12 PM - 8 PM
Open source & impact
7
Stars earned
29
Merged pull requests
Collaboration Quality
Co-authored commits
142
Middle Backend Developer
Confidence: Medium API Engineer
Backend engineer (senior-level) focused on document-grounded retrieval and API orchestration for RAG systems. The strongest proven skill is building and integrating a production-minded retrieval stack and service boundary (Python RAG HTTP/gRPC service + Go orchestration) as shown by api/http/app.py and grounded-horticulture_ru/server/rag_chat.go. There is limited public evidence of large-scale distributed systems work like sharding/consensus, measured performance tuning at scale, or an exhaustive cloud IaC/ops pipeline in the analyzed human-authored files.
API Design
6/10
How well APIs are designed
Consistent REST/gRPC surface and an internal retrieval service with auth separation, SSE streaming and admin endpoints; sound API design with conformance tests but no evidence of long-term versioning strategy or advanced idempotency patterns beyond session semantics.
Evidence
api/http/app.py: /rag/context, /domains, /admin/reindex endpoints and _rag_service_authorized/_admin_authorized
grounded-horticulture_ru/server/rag_chat.go: buildRAGLLMMessages, answerWithRAG and error/soft-fail handling
grounded-horticulture_ru/server/message_handlers.go: handleMessage, streaming vs multipart handling
Data Layer & Database
5/10
Working with databases
Clear, tested DB access patterns and sensible migration tooling; transactional migration recording is present in some modules but schema-evolution history is limited in the analyzed human-authored files.
Evidence
grounded-horticulture_ru/server/postgres_store.go: newChatStore, runAllMigrations, CreateSession, HistoryForLLM
grounded-horticulture_ru/server/postgres_store.go: careful NULL helpers (nullIfEmpty/nullToken/nullConfidence) and ownership SQL checks
Scalability & Performance
5/10
Handling load and speed
Hybrid retrieval, embedding caching and parallel evaluation tooling show attention to performance and scale, but there are no published load-test results or fine-grained, measured tuning artifacts in the human-authored files.
Evidence
rag/vector_store.py: hybrid search logic (Chroma + BM25 RRF, rerank, FETCH_K tuning)
scripts/run_rag_eval.py: ThreadPoolExecutor parallelism, in-process mode and timeout handling for retrieval smoke/eval
System Architecture
5/10
Overall system structure
Deliberate multi-process architecture (Go orchestrator, Python retrieval service, separate agent/gateway projects) with modular backends and admin/tenant boundaries; decomposition is pragmatic rather than exotic.
Evidence
api/http/app.py: Python RAG service as an internal microservice (HTTP/gRPC retriever)
grounded-horticulture_ru/server/rag_chat.go: Go orchestrator integrating Python RAG, LLM calls and verification path
Security & Auth
6/10
Protecting data and access
Good operational security hygiene at the API boundary: internal service tokens, admin secret checks, CORS guardrails and input validation; explicit production-secret validation exists and session ownership is enforced in DB queries.
Evidence
api/http/app.py: _require_production_secrets, _rag_service_authorized and explicit CORS policy checks
grounded-horticulture_ru/server/postgres_store.go: ReadImage validates tokens and session ownership queries (sessionOwned, SessionCropID)
Reliability & Observability
6/10
Stability and monitoring
Solid observability and resilience basics: metrics endpoints, retrieve latency recording, readiness checks, and startup/Postgres wait loops; graceful shutdown and request IDs are present in service glue code.
Evidence
api/http/app.py: /metrics and readiness endpoints, record_retrieve usage and structured logging of retrieve latencies
grounded-horticulture_ru/server/postgres_store.go: waitForPostgres with context-aware retries
Verified artifacts
Expertise
Go Development• Middle
Databases & Vector Storage• Middle
Python• Middle
Microservices & API Architecture• Middle
Industries
Artificial Intelligence• Middle
Farming & Agriculture• Middle
Technologies
Redis
Gunicorn
Docker• mentioned only
gRPC• mentioned only
Rest API• mentioned only
Recommendations
- Develop production RAG retrieval features (hybrid vector + BM25 + reranker), reindexing and embedding-caching improvements informed by real load tests.
- Build LLM orchestration and verified-answer pipelines (guardrails integration, streaming completions and SSE fallbacks) for enterprise APIs and SDKs.
- Implement hardened deployment and observability playbooks (Helm values for secrets, network policies, Prometheus/OTel dashboards and release gating) to convert demos into production.
- Extend DB migration history and add migration tests and idempotent deploy hooks to make schema evolution safer for multi-tenant installs.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle DevOps Engineer
Confidence: Medium CI/CD Engineer
Senior DevOps/Platform engineer specializing in CI/CD and release automation for LLM-powered RAG platforms. The strongest proven skill is pipeline and release engineering, evidenced by .github/workflows/ci.yml and release.yml with evaluation gates, Trivy scans, cosign signing and SBOM generation. Public code shows robust retrieval/agent services and deployment references, but there is limited evidence of hardened multi-environment Terraform remote state, documented SLOs or staged canary deploys in the human-authored artifacts.
CI/CD Pipelines
8/10
Automated build and deploy
Strong CI/CD and release engineering: multi-job GitHub Actions with test/scan gates, embedding cache and retrieval gates, image scanning, SBOM generation and cosign signing in release workflow.
Evidence
grounded-llm/.github/workflows/ci.yml: extensive CI jobs (python-test, eval-retrieval-gate, smoke-api, docker-build, helm-lint)
grounded-llm/.github/workflows/release.yml: cosign signing, Trivy scans, SBOM generation and CI gate step that requires CI success
Infrastructure as Code
4/10
Managing servers with code
Infrastructure as Code is present (Terraform reference modules and Helm chart), but reference-level rather than a hardened multi-env, remote-state, or complex state-migration practice.
Evidence
grounded-llm/deploy/terraform/aws/reference/main.tf: AWS reference Terraform stack for VPC, RDS, ECS task definitions
grounded-llm/deploy/README.md: Helm chart + Terraform reference guidance and values-prod.example overlay
Containerization & Orchestration
6/10
Working with containers
Containerization and orchestration practices are well represented: multi-stage Dockerfiles, healthchecks, Docker Compose, Helm chart lint/template in CI, and runtime configs for local LLMs and gRPC; some production k8s hardening and PDB mentions but limited evidence of tuned pod probes/affinity in the human-authored files shown.
Evidence
grounded-llm/docker-compose.yml: healthchecks, service ports, volumes and restart policies for server, python, postgres, redis, vllm/ollama profiles
mcp-gateway/Dockerfile: Go multi-stage build + nonroot runtime; mcp-gateway/docker-compose.yml: Jaeger and service wiring; CI helm-lint job references deploy/helm/grounded-llm
Observability & Monitoring
6/10
Watching system health
Observability and monitoring are embedded: Prometheus text endpoints, metrics instrumentation, OpenTelemetry traces (Jaeger), and CloudWatch log group in Terraform. No evidence of SLO/burn-rate alerting rules in code shown.
Evidence
grounded-llm/api/http/app.py: /metrics endpoint that emits Prometheus exposition and record_retrieve integration
mcp-gateway/docker-compose.yml and internal/otelx/otel.go: Jaeger OTLP + OpenTelemetry traces and metrics in proxy and registry code
Reliability & Incident Response
5/10
Keeping systems up
Reliability practices present: DB migration orchestration and idempotent migration tracking, startup readiness loops, graceful shutdown, smoke/backup CI jobs, and reindex locking. There are no explicit runbooks or chaos experiments in the human-authored files examined.
Evidence
grounded-llm/server/internal/store/postgres_store.go: RunAllMigrations with schema_migrations tracking and transactional apply
grounded-llm/.github/workflows/ci.yml: smoke-api and backup-smoke jobs; grounded-agent/cmd/server/main.go: graceful HTTP server shutdown
Cloud & Cost Optimization
3/10
Smart use of the cloud
Cloud and cost signals exist (ECS/Terraform task sizing, token cost estimate in gateway), but no advanced autoscaling, spot/eviction strategies, or documented rightsizing/cost-measurement artifacts in human-authored code.
Evidence
grounded-llm/deploy/terraform/aws/reference/main.tf: ECS task_definitions with cpu/memory settings and RDS config
mcp-gateway/internal/proxy/handler.go: estimateTokenCost used for per-tenant tool cost accounting and metrics
Expertise
CI/CD & Pipeline Automation• Middle
Infrastructure as Code• Middle
Kubernetes Ecosystem• Middle
Observability & Monitoring• Middle
Industries
Artificial Intelligence• Middle
Farming & Agriculture• Middle
Technologies
Containers
IaC
OpenTelemetry
Prometheus
Kubernetes
Docker• mentioned only
gRPC• mentioned only
Ollama• mentioned only
Rest API• mentioned only
vLLM• mentioned only
Recommendations
- Lead CI/CD and release automation for production LLM platforms, including image signing, SBOMs and gated retrieval/quality checks (use the existing .github/workflows/release.yml and ci.yml as a base).
- Implement and maintain Helm charts and reference Terraform stacks for staging/production (extend deploy/terraform and deploy/helm with remote state, secrets pipelines and PDBs).
- Build observability playbooks and SLO-driven alerting using existing metrics and traces (api/http/app.py metrics + Jaeger/OTel hooks in mcp-gateway).
- Develop reliable reindex/drift workflows and migration automation around the RAG vector stores, leveraging the existing reindex and migration tracking code.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
AI Workflow