Skip to content
Lorik Kastrati

Hello, I’m

Lorik Kastrati

Software Engineer · AI Engineer

6+ years building scalable backend and distributed systems with .NET, now building production AI applications with LLMs, RAG, AI agents, and automation.

Engineering
C#.NETASP.NET CoreMicroservicesDistributed Systems
AI
PythonLLMsRAGAI AgentsLangChainOpenAIVector Databases
Lorik Kastrati portrait

Prishtinë, Kosovo

About

I build production software from the systems layer up.

From distributed .NET services and event-driven architectures to AI applications powered by LLMs, retrieval, agents, and automation. I’ve spent six years shipping C# backends for platforms serving up to 20M users — and I bring that same rigor to AI: the model is one component, and the system around it is what makes it work in production.

Systems foundation

6+ years of production backend

  • Distributed systems
  • Microservices
  • Event-driven architecture
  • APIs
  • Databases
  • Messaging
  • Cloud infrastructure
  • Observability
AI specialization

LLM-powered applications

  • LLM applications
  • OpenAI
  • RAG
  • Embeddings & vector search
  • AI agents & tool calling
  • Structured outputs
  • OCR
  • AI automation & workflows
Currently Software Engineer at StarLabs
Years
0+

Shipping production backend systems

Users
0M

Reached by platforms I've contributed to

Services
0+

Microservices owned end-to-end

AI projects
0

LLM, RAG, and agent systems built end-to-end

Experience

Where I’ve been shipping.

6+ years as a software engineer on production backend systems — with AI engineering as my growing specialization.

Professional Experience
2021 – Present

StarLabs

Software Engineer

Backend microservices and integrations for high-traffic regulated platforms serving brands with up to 20M users.

  • Built and shipped C# / .NET microservices for Player Account Management (PAM) and third-party partner integrations, using Clean Architecture and event-driven design.
  • Used Microsoft Orleans for distributed state across a virtual actor cluster, eliminating shared-database contention for high-frequency wallet operations.
  • Implemented async messaging with Kafka and RabbitMQ (Outbox pattern via MassTransit) for reliable delivery and eventual consistency across services.
  • Cut hot-path latency with Redis caching and targeted PostgreSQL query and index work.
  • Owned CI/CD on TeamCity → Octopus → ArgoCD to Kubernetes, with structured Serilog logging and Kibana dashboards for production observability.
  • Applied CQRS (MediatR), Repository, and Unit-of-Work patterns; wrote unit, component, and automation tests across owned services.
.NETC#Microsoft OrleansKafkaRabbitMQgRPCPostgreSQLRedisKubernetes
2020 – 2021

Qonterra

Software Developer

REST APIs for a real-estate finance and analytics platform.

  • Built .NET REST APIs powering investment modeling and portfolio reporting.
  • Implemented financial calculation endpoints for IRR and other valuation / cash-flow methods used by analysts.
  • Contributed JavaScript front-end work to surface analytics in the client UI.
.NETC#JavaScriptRESTSQL
AI Engineering · Independent Projects
Ongoing

AI Engineering

Independent Projects

AI applications and systems built outside my primary employment, applying the same production standards to LLM-powered software.

  • Shipped AccountantAPI: invoice capture over Telegram, OpenAI Vision extraction into structured data, and a multi-tenant .NET 10 backend.
  • Built a RAG knowledge assistant from first principles — chunking, embeddings, Pinecone vector search, and grounded generation.
  • Designed tool-calling agents and a registry-driven multi-agent platform with process isolation and streamed output.
  • Added AI layers to full products: lead scoring and transcription in Sales Copilot, research and creative generation in Ad Studio.
PythonOpenAILangChainRAGPineconeAI Agentsn8n
Technology

The stack, by discipline.

Highlighted tools are the ones I reach for daily and know in depth.

Software Engineering

Six years of production backend work — the foundation everything else sits on.

  • C#
  • .NET 8 / .NET 10
  • ASP.NET Core
  • Microservices
  • Distributed Systems
  • Clean Architecture
  • DDD
  • CQRS
  • Event-Driven Architecture
  • REST
  • gRPC

AI Engineering

LLM applications, retrieval, and agents — built as real software, not just prompts.

  • Python
  • LLMs
  • OpenAI
  • LangChain
  • RAG
  • AI Agents
  • Vector Databases
  • LangGraph
  • Embeddings
  • Tool Calling
  • Structured Outputs
  • Prompt Engineering
  • AI Automation
  • OCR

Data & Infrastructure

Storage, messaging, and the operational tooling that keeps systems observable.

  • PostgreSQL
  • SQL Server
  • Redis
  • Elasticsearch
  • RabbitMQ
  • Kafka
  • Docker
  • Kubernetes
  • Azure
  • OpenTelemetry
  • Prometheus
  • Grafana
  • Jaeger
Specialization

AI Engineering

Building practical AI systems on top of strong software engineering foundations.

01

LLM Applications

Typed, validated LLM calls wired into real backends and user flows.

  • OpenAI
  • LLMs
  • Prompt Engineering
  • Structured Outputs
02

Retrieval

Grounding models in your own data with chunking, embeddings, and vector search.

  • Embeddings
  • RAG
  • Vector Databases
  • Semantic Search
03

Agents

Multi-step agents that pick tools, act, and reason over the results.

  • LangChain
  • LangGraph
  • Tool Calling
  • AI Agents
04

Automation

Document understanding and AI workflows connected to business systems.

  • OCR
  • n8n
  • AI Workflows
  • API Integrations
Projects

Things I’ve built.

AI systems first, then the backend engineering they're built on.

01

AI Engineering

LLM applications, retrieval, agents, and AI automation — designed as complete systems.

Featured case study
AI-Powered Accounting Platform

AccountantAPI

Multi-tenant accounting platform for Kosovo SMBs. Users photograph an invoice in Telegram; an OpenAI Vision pipeline extracts tax IDs, amounts, line items, and dates into a validated schema, and a 40+ node n8n workflow posts the result to a .NET 10 API — no manual data entry.

.NET 10OpenAI VisionOCRn8nPostgreSQLTelegram BotDocker
How it works
  1. 01Telegram
  2. 02Invoice image
  3. 03OCR / OpenAI Vision
  4. 04Structured invoice data
  5. 05n8n workflow
  6. 06.NET API
  7. 07Accounting database
AI layer
  • OpenAI Vision + OCR for invoice understanding
  • Structured outputs mapped to a strict invoice schema
  • AI workflow automation orchestrated in n8n
Engineering layer
  • .NET 10 backend with multi-tenant data isolation
  • Telegram bot as the capture channel
  • PostgreSQL persistence, Dockerized deployment
Architecture
  1. 01Documents
  2. 02Chunking
  3. 03Embeddings
  4. 04Pinecone
  5. 05Similarity search
  6. 06Retrieved context
  7. 07LLM
  8. 08Answer
Retrieval-Augmented Generation

RAG Knowledge Assistant

A question-answering assistant over private documents, built from the fundamentals up: document chunking, OpenAI embeddings, vector storage in Pinecone, similarity search, and context injection into the LLM prompt so answers stay grounded in the source material.

  • Chunking strategy tuned for retrieval quality
  • Embeddings + vector storage in Pinecone
  • Top-k retrieval with context injection
  • Grounded LLM generation
PythonLangChainOpenAI EmbeddingsPineconeVector SearchLLM
Architecture
  1. 01User request
  2. 02LLM
  3. 03Select tool
  4. 04Execute tool
  5. 05Tool result
  6. 06LLM
  7. 07Final response
Agent Architecture

LLM Tool-Calling Agent

An agent that turns a natural-language request into actions: the LLM selects a tool, the runtime executes it, and the result is fed back for the model to reason over — looping until it can produce a final answer. Tool inputs and outputs are typed with Pydantic for reliable, structured responses.

  • OpenAI tool calling via LangChain
  • Pydantic-typed tool schemas and structured outputs
  • Multi-step reasoning loop over tool results
PythonOpenAILangChainTool CallingPydanticStructured Outputs
Multi-Agent Automation Platform

AI Agents Workforce

A registry-driven set of specialized agents — product research, lead prospecting, outreach drafting, ad scripting, clip posting, storefront building, and job search. A single agents.json powers both a Streamlit dashboard and a Telegram bot; every agent runs isolated in its own environment as a subprocess with live-streamed output.

  • One registry drives the dashboard and the Telegram bot
  • Process-isolated agents with streamed output
  • Agents chain into each other (prospect → pitch → draft)
PythonClaude Agent SDKAI AgentsStreamlitTelegram Botuv
AI Product Content & Advertising Platform

Ad Studio

Paste a product link and get a consistent multi-angle product lookbook, TikTok/Meta ad scripts and creatives, or a generated Shopify storefront — all in the shop's house style.

AI layer
  • AI agents for product research and ad generation
  • LLM workflows for scripts and copy
  • Image generation for product shots and creatives
Engineering layer
  • FastAPI backend orchestrating the agent as a subprocess
  • Results streamed to a Vite + React front end
PythonFastAPIClaude Agent SDKImage GenerationReact
AI Sales & CRM Platform

Sales Copilot

A sales platform that takes prospects from first contact to closed deal, with AI built into the workflow rather than bolted on — scoring leads, driving outreach, and transcribing calls in real time.

AI layer
  • AI-assisted lead scoring
  • AI-driven sales workflows and automated outreach
  • Live call transcription with Deepgram
Engineering layer
  • .NET 8 layered backend (Domain / Application / Infrastructure / API)
  • Hangfire background processing, PostgreSQL
  • SMS/email threads, Telnyx WebRTC calling, Next.js frontend
.NET 8DeepgramHangfirePostgreSQLTelnyx WebRTCNext.js
02

Software Engineering

Production backend and distributed systems work in .NET.

Multi-Tenant Payment & Billing Platform

PayAPI

Centralized billing platform serving multiple SaaS products from one API. End-to-end Stripe integration with PaymentIntents, subscriptions, and invoices, driven by webhooks and made safe to retry with idempotency keys.

  • Stripe PaymentIntents, subscriptions, webhooks
  • Idempotent payment operations
  • API keys for services, JWT + RBAC for admins
  • Tenant-isolated data model
.NET 8PostgreSQLStripeWebhooksJWT / RBACMulti-tenancy
Distributed Microservices Platform

ELibrary

Books, Users, Orders, and Consumer services behind a YARP gateway, communicating asynchronously over RabbitMQ with MassTransit. Elasticsearch powers catalog search, and the whole system is traced and measured end to end.

  • Event-driven services over RabbitMQ + MassTransit
  • YARP API gateway, Elasticsearch search
  • OpenTelemetry → Prometheus, Grafana, Jaeger
.NET 8RabbitMQMassTransitElasticsearchYARPOpenTelemetryGrafanaJaegerDocker
Get in touch

Let’s build something together.

Best way to reach me is email. I read everything and reply within a day — happy to talk Software Engineer and AI Engineer roles, backend architecture, or production AI systems.