Едуард

Eduard Chirva
Python Backend Developer · AI / LLM Engineer
Async Python backend · LLM integration (cloud & self-hosted) · Agentic systems · RAG

[відкрити контакти](див. вище в блоці «контактна інформація») · t.me/qvartzgolden20 · github.com/qvartzggolden01 · Ukraine · Remote · Ukrainian (native) ·
Russian (fluent) · English B1

Python engineer with two years hands-on building backend and applied-AI systems: async backends, LLM
integrations across cloud and self-hosted models, agentic systems, RAG pipelines, and computer vision. I write the
code and design the architecture around it, owning every decision and using AI tooling to move faster while
staying able to explain and defend the code line by line.
Before development: around six years in IT from system administrator to Head of IT (team of up to 30, a 200
person company, infrastructure built from scratch, budgets, ITIL processes). My strength is whole-system depth,
from infrastructure and networking up to the API and the agents on top.

PROJECTS

COMMERCIAL WORK

Lead-Processing & Automation Platform Commercial · 3 mo · Solo

Collapsed a manual spreadsheet-and-chat workflow into a single platform. Node.js + Python over a custom IPC
protocol JSON-lines on stdin/stdout), a mobile-first multi-role UI, and a messaging-integration layer with rate-limit
and reconnection handling.
150+ active users, 130K+ records; 6× revenue over operating cost (≈$10K opex → $60K ; runs on 2 CPU / 4 GB RAM
Stack: Node.js, Express, Python, PostgreSQL, Docker, supervisord, WebSocket, OpenAI API

Computer-Vision Automation Pipeline Commercial · 4 mo · Solo

End-to-end applied-ML automation across three layers. Perception: a custom-trained YOLO model (own dataset,
augmentation, transfer learning, ONNX / TensorRT export, active-learning loop). Decision: a weighted action
orchestrator. Execution: programmatic device control.
500+ concurrent device instances, 24/7; replaced thousands of manual man-hours per month
Stack: Python, YOLO, OpenCV, ONNX / TensorRT, LLM API

Conversational LLM Platform Commercial · In production

Conversational platform on LangChain / LangGraph with a Weaviate vector database SQLite fallback), automatic
language detection CLD2 , voice transcription, and structured background workers for reporting and export. Built
as an observable LLM pipeline rather than one-off prompt scripting.
Stack: Python 3.12, FastAPI, LangChain / LangGraph, Weaviate, PostgreSQL, Docker

Transactional Marketplace with Escrow Commercial · 4 wk · Solo

Multi-component Telegram-based marketplace: escrow with an internal wallet, auto-deposit and manual
withdrawal approval, a community layer, and a Mini App for counterparty checks. Anti-abuse built in: protection
against malicious-approval schemes and transaction-integrity safeguards.
Paid for itself within 2 weeks of launch
Stack: Python, JavaScript, PostgreSQL, Docker, Telegram Mini App, Bot API

Campaign Analytics Bot Commercial · Delivered

Event-driven analytics automation: an FSM funnel, deeplink parameter parsing, idempotent server-to-server
postbacks, a structured event stream for BI, payload-based A/B testing, and multi-language + GEO-dependent
currency.
Stack: Python, aiogram, SQLite, httpx, FSM, event logging, i18n

PERSONAL & R&D

Self-Hosted LLM Serving Infrastructure + Autonomous Agent R&D · In use

Self-hosted LLM platform on a single consumer 24 GB GPU with three serving modes: an always-on pool, a hot-
swap manager (asyncio.Lock), and a layer-split GPU RAM mode that runs 70B models on one GPU. Custom
FastAPI gateway with token auth and an 8-stage request pipeline; hybrid RAG LanceDB BM25, Reciprocal Rank
Fusion). Fully air-gapped. Includes a code agent Plan / Execute / Audit / Debug, 750 lines, minimal deps) with a
stdlib-only web UI 900 lines).
Stack: Python, FastAPI, llama.cpp CUDA , GGUF, LanceDB, BM25, asyncio

Async Multi-Source Data Engine Personal · Running 24/7

Long-running async backend: 18 concurrent loops under a supervisor pattern, a 12-stage data pipeline with 3
layer deduplication and 6-factor scoring, a fault-tolerant Redis wrapper, and cost-optimised routing across three
LLM tiers. Core scoring and aggregation logic implemented from scratch without heavy third-party libraries.
8,500 lines of Python across 36 files, built for continuous fault-tolerant operation
Stack: Python, asyncio, SQLAlchemy 2.0 (async), PostgreSQL, Redis, Docker, Claude API, JSON RPC

Async Request Classifier Public repo · Code sample

Async multi-source classifier written from scratch: semaphore-based rate limiting, exponential backoff, provider
abstraction via ABC, and Pydantic schemas. Public on GitHub as a hands-on Python sample.
Repo: github.com/qvartzggolden01/netpeak-request-classifier
Stack: Python, asyncio, Pydantic, ABC

NFC Attendance Tracker In-house · Daily use

USB NFC reader into a zero-dependency Python console app (stdlib only). State machine with no persistent state;
an append-only raw log as the source of truth plus a CSV with dynamic columns written atomically (temp +
rename). Deployment is copying the folder.
Stack: Python (stdlib only)

IN DESIGN / IN PROGRESS

Maritime-Risk RAG Platform In design

Retrieval-augmented system for maritime risk assessment where answers must be grounded and citable. Hybrid
retrieval combines dense BGE M3) and sparse BM25) signals fused with Reciprocal Rank Fusion over LanceDB,
so responses come from source documents with attribution. In design toward a first engagement.
Stack: Python, BGE M3, BM25, RRF, LanceDB

Local Voice + LLM Assistant In design · spec complete

End-to-end architecture for a fully local, real-time voice assistant under an on-prem constraint (call data cannot
leave the perimeter). Pipeline: telephony Kamailio / FreeSWITCH → VAD Silero → STT Parakeet) → local LLM
(SGLang + Qwen3 8B) with constrained decoding XGrammar) and PII redaction Presidio) → TTS Kokoro ,
grounded on hybrid RAG BGE M3 LanceDB . Complete technical specification produced.
Stack: SGLang, Qwen3 8B, Pipecat, Kamailio / FreeSWITCH, Parakeet, Kokoro, Silero VAD, XGrammar, Presidio, LanceDB

EXPERIENCE

IT System Administrator → Head of IT
Commercial & sales services group 200 staff) · 2020 2026

Joined as the company's sole system administrator and built the IT function from scratch as the business
scaled from a small office to a 200-person group.
Ran all core infrastructure (local networks, servers, RDP, VPN, SIP telephony, surveillance, end-user support)
and introduced ITIL-style processes as the team grew; led and coordinated a team of up to 30.
Owned budget-to-productivity optimisation: kept infrastructure efficient while scaling, then significantly cut RDP
and maintenance spend during a later cost-optimisation phase.
Over the last two years, moved personally into hands-on development (internal tools in Python, LLM integration,
computer vision), building working systems end to end while upskilling the team.

TECHNICAL SKILLS

Languages 2 yrs Python (primary), SQL, JavaScript / Node.js, TypeScript

Backend 2 yrs FastAPI, asyncio, aiohttp, httpx, Pydantic v2, SQLAlchemy 2.0 (async), REST APIs,
background workers, rate-limit handling, custom IPC, Express.js

AI / LLM 2 yrs LLM integration (cloud: OpenAI, Claude APIs; self-hosted: open models), agentic
systems (autonomous agents, LangChain, LangGraph), RAG BGE M3 BM25 RRF
hybrid retrieval, LanceDB, Weaviate), prompt engineering, constrained decoding, PII
redaction

LLM serving / self-hosting 1 2 yrs SGLang, llama.cpp CUDA ; quantization tradeoffs GGUF, FP8, AWQ INT4 , KV-cache
sizing & capacity planning, continuous batching, air-gapped deployment

ML / Computer Vision 2 yrs YOLO (custom training, transfer learning), OpenCV, ONNX / TensorRT, active-learning
pipelines

Data & Infrastructure 3 yrs PostgreSQL, Redis, SQLite, LanceDB, Weaviate; Docker / Compose, supervisord,
CI/CD

Networking / SysAdmin 6 yrs MikroTik, local networks, VPN, RDP, SIP telephony, server management

Methodology Design-first / spec-driven development, AI-augmented implementation while owning
and defending the code, ITIL-style IT processes

WORK FORMAT

Fully remote, any timezone. End-to-end delivery: architecture, implementation, deployment, production support. Comfortable with
ambiguity and able to turn a business problem into a working system. Strong fit for backend and applied-AI roles that value both
hands-on engineering and whole-system understanding.