adAndrei
Dashkevich.
INDEPENDENT AI ENGINEER / KRAKÓW, PL

Turning AI
into working
systems.

Helping teams build faster,
automate complex work and find accurate answers
in their own data.

From the first prototype
to reliable production.
SYSTEM ARCHITECTURE 01 / 03
⌘
01 / KNOWLEDGEThe right context
02 / ORCHESTRATIONIntelligent agents
✓
03 / EVALUATIONVerified outcomes
TOOLS.
NOT LIMITATIONS.
Claude Code⌘ Codex◈ Cursor⌁ OpenCodeLangGraph LangChain ∞
01 / EXPERTISEBuilt for your challenge. Aligned with your business.

Less busywork.
More possibilities.

Three areas of expertise. One goal:
AI that makes a practical difference.

01
AI-POWERED DEVELOPMENT

AI-powered development.
A complete engineering workflow.

I configure Claude Code, Codex, Cursor and OpenCode around your repository, team and workflow — from the brief to the release.

Claude CodeCodexCursorOpenCode
  • The full cycle: planning → code → review → testing → CI/CD and deployment.
  • Project rules, skills, MCP integrations and context management.
  • Token efficiency through model selection, caching and precise instructions.
  • Refactoring, removing unnecessary code and fixing vulnerabilities.
  • Quality gates, sandboxing and oversight of agent actions.
02
AGENTIC BUSINESS AUTOMATION

Agents that take tasks
all the way to completion.

I design complex agent systems for business workflows, with state, memory, tools and clearly defined autonomy.

LangGraphLangChainDeepAgents
  • Multi-step workflows and teams of specialized agents.
  • Integrations with CRMs, APIs, databases and internal services.
  • Human-in-the-loop approval for consequential decisions.
  • Persistent state, retries and recovery from failures.
  • Tracing, quality evaluation, cost control and latency monitoring.
03
RETRIEVAL-AUGMENTED GENERATION

Your company’s knowledge.
Answers you can trust.

I build RAG systems on your internal knowledge base, tuning every retrieval and generation step using evaluations on your data.

Dense + SparseBM25RerankingEvaluation
  • Vector, lexical (BM25) and hybrid retrieval.
  • Chunking, embedding and reranker selection, with fine-tuning on domain-specific data.
  • Query transformation: multi-query, HyDE, decomposition and query rewriting.
  • Context expansion: parent-child retrieval, neighboring chunks and metadata enrichment.
  • Source citations, groundedness checks and abstention when evidence is insufficient.
  • Document access control, index updates and prompt injection defenses.
DEEP DIVE / KNOWLEDGE SYSTEMS

Retrieve what matters.
Generate with precision.

Answer quality starts long before the prompt.
I engineer the entire path, from document to verified answer.

PROJECT TARGET
up to 96%*
accuracy on an agreed evaluation dataset
RETRIEVAL LABChoose a retrieval strategy ↓
03 / HYBRID RETRIEVAL

Meaning + exact matches.

Combine semantic retrieval and BM25, merge results through fusion and promote the most relevant passages with a reranker.

For mixed knowledge bases: prose, names, codes and terminology.
↳ User query
◈ Vector retrievalMeaning and context
⌗ BM25Terms and matches
≋ Fusion + Reranker→Context for the LLM ✓

Also available: multilingual retrieval, OCR for complex documents, tables, knowledge versioning and regression evaluations.

*96% is a potential project target, not a guarantee. Metrics, evaluation data and acceptance criteria are agreed in advance. Results depend on the data, domain and complexity of the questions.

02 / APPROACHA transparent process. Measurable quality.

From “what if”
to “it’s live.”

We start with your challenge.
We deliver a system your team can maintain.

01

Discover

Audit workflows, data and constraints. Define the use case, success metrics and the cost of errors.

Deliverable: a plan and quality criteria
02

Validate

Prototype on your data. Compare approaches and measure quality, latency and cost.

Deliverable: a PoC and impact assessment
03

Deploy

Integrations, testing, security and deployment. Prepare the system for real-world workloads.

Deliverable: a working solution
04 ✓

Improve

Monitoring, evaluations and feedback. Optimize quality, token usage and operating costs.

Deliverable: visibility and ongoing improvement
03 / ABOUT KRAKÓW, POLAND · REMOTE
ENGINEER FIRST. AI SECOND.

Andrei Dashkevich.

Bringing AI and engineering together.

I focus on systems that solve a specific business problem, with clear architecture, verifiable quality and controlled costs.

I help you choose an approach, build the solution and equip your team to keep it running — from a one-off audit to full implementation.

AI EngineeringAgentic SystemsProduction RAG
LET’S BUILD SOMETHING USEFUL

Where do
we start?

Tell me what you want to improve.
Let’s explore where AI can make a difference
and find the right first step.

A little about your project

Your email app will open with a draft. You review and send it. Form data is not stored on this site.