AI & LLM Engineering
RAG pipelines, prompt and context design, tool-calling agents, guardrails, evaluation harnesses and cost control.
RAGAgentsEvals
Last 2 years: AI engineering with Python & AWS
I build AI systems that survive production.
For the past two years my work has been AI engineering in Python on AWS — retrieval pipelines, LLM services and agent workflows that real businesses depend on. Behind that sits 20+ years of enterprise Java/Spring Boot architecture, which is exactly why the AI I ship is observable, secure and maintainable rather than a demo.
- 20+
- Years building enterprise backends
- 2
- Years focused on AI with Python & AWS
- 8+
- Enterprise systems delivered end-to-end
- ZK
- Official ZK & Keikai implementation partner
What I'm doing now
Two years deep in AI, Python and AWS
I moved my practice into AI because that is where the hard, interesting problems moved. The stack changed — Python instead of Java, Bedrock and Lambda instead of Tomcat — but the discipline did not. These systems still need schemas, tests, cost ceilings and a way to debug them at 2 a.m.
-
Retrieval-Augmented Generation
Document ingestion, chunking strategy, embeddings and vector search that returns grounded answers with citations — not confident guesses.
-
LLM services & agent workflows
Tool-calling agents and multi-step pipelines with guardrails, retries, timeouts and human-in-the-loop checkpoints where the stakes justify them.
-
AWS-native architecture
Bedrock, Lambda, API Gateway, S3, DynamoDB, OpenSearch, Step Functions and ECS/Fargate — provisioned as code, with IAM scoped properly.
-
Evaluation, cost & observability
Regression suites for prompts, token and latency budgets, structured tracing. You get to see what the model did and what it cost.
-
AI inside existing enterprise systems
The niche I own: adding AI to ERP, CRM and finance platforms that already work, without destabilising the system paying the bills.
# Grounded answers over enterprise documents
from fastapi import FastAPI, Depends
from pydantic import BaseModel, Field
app = FastAPI(title="Knowledge API")
class Query(BaseModel):
question: str = Field(min_length=3)
top_k: int = 6
class Answer(BaseModel):
text: str
citations: list[str]
cost_usd: float
@app.post("/ask", response_model=Answer)
async def ask(q: Query, svc=Depends(get_rag)):
# vector search -> rerank -> Bedrock
hits = await svc.retrieve(q.question, k=q.top_k)
return await svc.generate(
q.question,
context=hits,
guardrails=True,
max_cost_usd=0.05,
)
Python 3.12
FastAPI
Pydantic
RAG
LangChain
Embeddings
Agents
Bedrock
Lambda
SageMaker
OpenSearch
Step Functions
DynamoDB
S3
The through-line
Why an AI engineer with 20 years of Java is a feature, not a detour
Most AI projects do not fail on the model. They fail on integration, data access, permissions, cost and the fact that nobody can tell why the output changed. Those are enterprise backend problems, and I have been solving them since long before "AI engineer" was a job title.
I work either directly as an independent specialist, or as a sales and engagement manager representing Quontex when a project needs a full delivery team.
Since 2024 · Current focus
AI engineering — Python & AWS
Designing and shipping AI features: RAG over private document sets, LLM-backed services, agentic automation, and the evaluation and cost controls that make them safe to run. Python end to end, deployed on AWS.
LLM / RAG
Python
AWS
2021–2024 · Enterprise reporting
Keikai & advanced spreadsheet solutions
Five years integrating Keikai into enterprise applications to deliver spreadsheet-driven business solutions — custom reporting, financial analysis and Excel-grade interaction inside web systems.
Keikai
ZK 10
Spring Boot
Ongoing · Long-standing specialism
ZK Framework, version 3 through 10
Deep practical experience across every major ZK generation, including custom components and phased migrations. Recognised as an official ZK implementation partner for enterprise-grade delivery.
ZK 3–10
Custom components
Since the 2000s · Foundation
Java & Spring Boot architecture
20+ years of reliable, scalable, high-performance web applications: REST API design, fixing legacy and poorly performing backends, third-party integrations, and troubleshooting production systems that cannot go down.
Java
Spring Boot
Hibernate
SQL
At a glance
Alex Koziy at a glance
A plain summary of who I am and what I do, for people and machines in a hurry.
- Name
- Alexander (Alex) Koziy
- Current focus
- AI engineering with Python and AWS (last 2 years)
- Core experience
- 20+ years, Java and Spring Boot enterprise architecture
- AI stack
- Python, FastAPI, RAG, LLM agents, Amazon Bedrock, Lambda, OpenSearch
- Enterprise stack
- Java, Spring Boot, Hibernate, ZK 3–10, Keikai, SQL databases
- Credential
- Official ZK & Keikai implementation partner
- Industries
- Finance, government, industrial, transportation, real estate, ERP/CRM
- Engagement models
- Direct as an independent specialist, or a dedicated team via Quontex
- Clients & partners
- PwC, Stratifytech, Dynamic Inventory
- Location
- Remote, working with clients worldwide
- Availability
- Open to new engagements; replies within 1 business day
- Contact
- info@alexkoziy.com
Expertise
Four pillars I can be hired for
Pillar one is where I spend my time today. Pillars three and four are why clients trust me with pillar one.
Python Backend
Async FastAPI services, Pydantic schemas, clean service/repository layering, typed code and tests that actually run in CI.
FastAPIPydanticpytest
AWS Cloud
Serverless and container architecture, Bedrock and SageMaker integration, infrastructure as code, least-privilege IAM.
BedrockLambdaFargate
Java, Spring Boot & ZK
Legacy modernization, JVM profiling and performance work, REST integration, and ZK/Keikai enterprise reporting.
Spring BootZK 3–10Keikai
Official ZK & Keikai implementation partner
Recognised by ZK for delivering enterprise-grade solutions with ZK and Keikai — a credential that still matters when AI has to live inside an existing enterprise UI.
Selected work
Systems in production, not prototypes
Enterprise platforms across finance, government, industry and logistics — several still running years after delivery. Full details on the case studies page.
Financial Planning & Analysis
Consolidates complex financial and operational data so organizations can generate custom reports, visualize trends and make confident decisions across business units.
- Stack
- Spring Boot, ZK, Keikai, AWS
- Impact
- Keikai integration enabled custom reporting and streamlined financial analysis
Dynamic Public Safety
Law enforcement software built from the ground up for complex criminal justice workflows — case management, reporting and multi-agency data exchange.
- Stack
- Java, ZK, MySQL, Hibernate, Tomcat
- Impact
- Improved UI responsiveness and scalability for statewide deployments
Interactive Fleet Map
Real-time vehicle tracking that visualizes live locations and historical routes on an interactive map, so transport operators can monitor fleet activity as it happens.
- Stack
- Java, ZK, OpenLayers, PostGIS
- Impact
- Improved fleet visibility and operational control
How to work with me
Two ways to engage
Hire me directly as a specialist, or bring in a full delivery team through Quontex. Same person accountable either way.
Individual expertise
AI engineer · Python · AWS · Java
Work directly with me as a senior specialist. Best when you need judgement and hands-on delivery rather than headcount.
Best suited for
- Adding AI features to an existing product or platform
- Proving out an AI use case before committing a team
- Existing applications needing expert backend support
- Architecture design, modernization and performance work
- Long-term maintenance and scalability ownership
Dedicated teams & project delivery
Via Quontex · Presales & engagement management
I work with clients on behalf of Quontex to structure and launch dedicated development teams for enterprise projects, covering presales, engagement setup, delivery coordination and long-term partnership management.
Best suited for
- New enterprise projects and greenfield development
- Rapidly scaling delivery capacity with a dedicated team
- Full-cycle delivery: discovery, build, long-term support
- Long-term product development partnerships
- CTOs and product owners who need predictable delivery
Clients & partners
Companies I've worked with
Enterprise and long-term engagements.
FAQ
Frequently asked questions
The questions clients and recruiters ask most often, answered directly.
Who is Alex Koziy?
Alex Koziy (Alexander Koziy) is a software engineer who builds AI systems in Python on Amazon Web Services. He has more than 20 years of experience in enterprise Java and Spring Boot architecture, and for the last two years his work has focused on AI engineering. He is a recognised official ZK and Keikai implementation partner and works with clients worldwide on a remote basis.
What is Alex Koziy working on now?
For the past two years Alex Koziy has worked on AI engineering using Python and AWS. That includes retrieval-augmented generation (RAG) over private enterprise document sets, LLM-backed services, agent workflows that call real business systems, and the evaluation and cost controls needed to run them safely in production.
What AI services does Alex Koziy provide?
He provides retrieval-augmented generation pipelines including document ingestion, chunking strategy, embeddings and vector search; LLM services and tool-calling agent workflows with guardrails, retries and human-in-the-loop checkpoints; AWS-native AI architecture; evaluation harnesses with token, latency and spend budgets; and integration of AI into existing ERP, CRM and finance platforms.
Which AWS services does he use to build AI systems?
Amazon Bedrock for model access, AWS Lambda and Amazon ECS/Fargate for compute, Amazon API Gateway for interfaces, Amazon S3 for document storage, Amazon DynamoDB for state, Amazon OpenSearch Service for vector search, AWS Step Functions for orchestration, and Amazon SageMaker where custom models are needed. Infrastructure is provisioned as code with least-privilege IAM.
Does Alex Koziy still work with Java and Spring Boot?
Yes. Java and Spring Boot remain part of his practice, particularly for legacy modernization, REST API development, JVM profiling and performance optimization, and for adding AI capabilities to existing Java enterprise systems without destabilising them. The AI work is an addition to that foundation, not a replacement for it.
What is his experience with ZK Framework and Keikai?
He has worked with the ZK Framework across every major generation from version 3 through version 10, including custom component development and phased migrations. Over the last five years he has integrated Keikai into enterprise applications to deliver spreadsheet-based business solutions such as custom reporting and financial analysis. He is a recognised official ZK and Keikai implementation partner.
What industries has Alex Koziy built software for?
Finance and financial planning and analysis, government and public safety, industrial and energy management, transportation and fleet tracking, real estate, and enterprise ERP, CRM and e-commerce systems. Delivered platforms include financial planning and analysis tools, law enforcement case management, contract management, inventory management, energy monitoring and real-time fleet mapping.
How can you hire Alex Koziy?
There are two engagement models. You can hire him directly as an independent senior specialist for AI, Python, AWS or Java work, which suits adding AI features to an existing product, architecture and modernization work, and long-term maintenance. Alternatively he can arrange a dedicated development team through Quontex, which suits greenfield enterprise projects and rapidly scaling delivery capacity.
Is Alex Koziy available for remote work?
Yes. He works remotely with clients worldwide and is currently available for new engagements. He typically responds to enquiries within one business day.
How do you contact Alex Koziy?
By email at info@alexkoziy.com, by booking a free 30-minute consultation call through the scheduling link on alexkoziy.com, or through his LinkedIn or Upwork profiles. Email and a booked call are the fastest routes.
Contact
Let's talk about your AI or backend project
Tell me what you're building or what's breaking. If it isn't something I should take on, I'll say so and point you somewhere better.
Book a 30-minute callFree consultation · no obligation
info@alexkoziy.comDirect email — fastest written route
LinkedInProfessional background & network
UpworkContract through an escrow platform
Freelancer.comBid-based projects and contests
FiverrFixed-scope packages
Fastest response via call or email. I usually reply within one business day.
Thank you for your message
I'll get back to you shortly — usually within one business day.
Have an AI idea that needs to survive contact with production?
That's the gap I work in — between an impressive demo and a system your business can actually depend on.