Kirill

Savchenko Kirill
AI Developer
Odesa, Ukraine | [open contact info](look above in the "contact info" section) | [open contact info](look above in the "contact info" section)
GitHub ks-77 | Telegram @kirill_sv7

Professional Summary⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
Full-stack developer focused on backend engineering and applied AI. Builds web applications end to end, from user
interfaces and backend APIs to data integrations and deployment. Works with Django, Flask, and FastAPI, choosing
frameworks and tools that fit each project. Develops AI agents and integrates LLMs into business workflows. Attends
conferences on applied AI.
Experience
Singularika | Full-time, remote
Chief Technology Officer (CTO) | Solution Architect | Lead AI Developer December 2025 – November 2026
Lead product architecture and technology strategy. Select the technology stack and oversee technical decisions from
design through implementation of core product features.
Python Developer December 2024 – December 2025
NDA | Part-time, remote
Python Developer May 2025 – February 2026
Contributed to the development of analytics tools for an education provider focused on on-chain crypto. Integrated AI
tools into the sales team's workflows.

Projects ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
Aurora | Multi-Tenant AI Assistant Platform
Overview — Platform for configuring AI assistants across business workflows at a Ukrainian retailer.
Capabilities — Teams can tailor assistants by selecting models, agents, prompts, and knowledge bases.
Key Contributions
• Built a multi-tenant backend that isolates API credentials, datasets, and object storage.
• Orchestrated supervisor and specialist agents in LangGraph, combining tenant-scoped hybrid search with
prompt versioning.
• Built asynchronous document ingestion and batch indexing. Added response streaming and tracking for model
usage and cost.
Tech Stack — Python, FastAPI, LangGraph, PostgreSQL, Qdrant, Celery, Redis, MinIO, Google Drive API,
OpenRouter, Langfuse

Atmosfera | Technical Support Assistant
Overview — AI assistant for a distributor of solar and backup power equipment.
Capabilities — Supports equipment selection, compatibility checks, and troubleshooting in a single conversation.
Key Contributions
• Implemented intent-based routing in LangGraph across catalog queries, manual search, and specialist support
workflows.
• Built multi-turn equipment identification and used the selected model to filter hybrid retrieval over manufacturer
documentation.
• Added a keyword search fallback that retains equipment filters during embedding service overload. Added
tracing and regression tests for retrieval recovery.
Tech Stack — Python, FastAPI, LangGraph, LangChain, SQLAlchemy, MySQL, Qdrant, OpenRouter, Langfuse

Projects continued
Kernel | Employee Learning and Knowledge Assistant
Overview — Internal AI assistant for employee onboarding, training, and access to company information.
Capabilities — Employees can find learning materials, get answers to common questions, and look up employee
records in natural language.
Key Contributions
• Built document ingestion and added hybrid search with filters based on employee attributes to retrieve relevant
internal materials.
• Implemented read-only SQL lookups with query validation, a table allowlist, and result sanitization.
• Updated tracing and dataset-based evaluations to compare assistant responses with reference answers.
Tech Stack — Python, FastAPI, LangGraph, MySQL, Qdrant, Pinecone, OpenAI API, Langfuse

Hard skills ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
Python backend
• Python, FastAPI, Django, Django REST Framework (DRF), Flask, SQLAlchemy, Alembic, REST APIs, async
programming, background tasks, microservices, event-driven architecture, WebSockets, Server-Sent Events
(SSE), gRPC
Data and storage
• PostgreSQL, MySQL, Redis, ClickHouse, MinIO, Qdrant, SQLite
LLM integration and inference
• OpenAI API, Anthropic Claude API, Google Gemini API, DeepSeek API, prompt engineering, structured outputs,
function calling, tool calling, Ollama, Hugging Face, local and self-hosted LLM deployment, inference cost
optimization
RAG and search
• Elasticsearch, full-text search, semantic and hybrid retrieval, vector search and indexing, embeddings, reranking,
metadata filtering, query routing, Self-RAG, context compression, citation extraction, retrieval pipelines and
evaluation, knowledge bases, semantic caching
Document intelligence
• OCR, PDF and Office document parsing, layout analysis, extraction of document structure, tables and images,
chunking strategies, metadata enrichment, document classification and indexing, asynchronous document
processing, knowledge base ingestion pipelines
AI agents and automation
• LangGraph, MCP (Model Context Protocol), stateful agents, multi-agent orchestration and collaboration,
tool-based agents, SQL and data agents, human-in-the-loop workflows, agent memory, planning and execution
loops, workflow automation with n8n, external API and internal business system integrations
LLMOps
• Langfuse, LangSmith, LLM call tracing, agent observability, monitoring of token usage, latency and cost, prompt
versioning, dataset management, LLM regression testing, production analytics
Infrastructure and DevOps
• Docker, Docker Compose, Nginx, Linux, AWS, CI/CD, GitLab CI/CD, GitHub Actions, deployment automation
and production AI service configuration
Developer tools
• Git, GitHub, GitLab, Postman, Swagger, OpenAPI, PyCharm, Jupyter

Soft skills ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
• Problem framing: Break ambiguous requirements into concrete user scenarios, edge cases, and acceptance
criteria that can be tested.
• Technical communication: Explain what an assistant can answer, which sources it uses, and where it needs
clarification in terms that technical and business colleagues can use.
• Cross-functional collaboration: Align with colleagues on domain terminology and the data an assistant needs
for accurate answers, such as equipment models or internal policies.
• Accountability — Take ownership of commitments, flag blockers early, and follow through to completion.