Maulenov

Rodion Maulenov
AI AUTOMATION & BACKEND ENGINEER
Kyiv, Ukraine · [open contact info](look above in the "contact info" section) · [open contact info](look above in the "contact info" section)

SUMMARY

Backend and AI-automation engineer with an unusual combination: I run client acquisition and advertising for a clinic group's Asian
market, and I am also the engineer who automates that work. Sole architect of a ~100k-LOC Python platform — six internal
products, a Mini App, telephony and computer-vision integrations, and an autonomous self-repair engine — deployed and operated
on AWS. I build systems that turn unstructured input (calls, photos, PDFs, video, free text) into structured, queryable data with
LLMs in the loop, and I know exactly what the marketing side needs from them, because I am the one who needs it.

CORE SKILLS

Languages Python 3.13 (expert, async-first), TypeScript / JavaScript ES2022, SQL, Bash
Backend Framework-agnostic; production experience with FastAPI, aiogram 3.x, asyncpg, Pydantic v2, Dishka (DI),
uvicorn/uvloop, structlog. Comfortable picking up any backend stack — Django, Flask, Litestar, Node/Express.
Frontend Angular, TypeScript, vanilla ES2022 modules, responsive mobile/WebView UI, Biome/ESLint, component
testing
Data PostgreSQL 17 (schema design, indexing, migrations, event sourcing), Redis, ETL & idempotent pipelines,
JSONB, object storage
Cloud AWS (production workloads on EC2, S3, IAM, registry-based image delivery) · GCP (Vertex AI, Cloud Storage,
service-account & quota management)
AI / LLM Claude (Agent SDK, CLI process pools, MCP tool servers), Google Gemini Vision on Vertex AI, prompt & agent
design, structured extraction, LLM-as-judge evaluation, multi-agent orchestration, cost & rate-limit budgeting
Marketing Meta & Google Ads run hands-on (~$6.5k/month), lead capture and attribution, funnel instrumentation,
campaign budget reallocation by measured channel performance
Automation Webhooks, REST APIs, n8n, Telegram Bot API + self-hosted Bot API server, MTProto userbot automation,
telephony APIs, scheduled jobs (APScheduler, systemd timers, cron)
Infrastructure Docker & Compose (multi-stage, BuildKit, hardened runtime), GitHub Actions CI/CD, Caddy + automatic TLS,
journald log pipelines, backup & restore drills
Quality pytest (6,000+ tests, parallel xdist, per-worker Postgres), mypy strict, ruff, Biome, custom mock Telegram API
server for end-to-end flows

EXPERIENCE

Biotexcom — fertility clinic group 2016 – Present

Software Engineer — AI Automation 2021 – Present
Sole engineer and architect of the company's internal automation platform: a modular Python monolith running six product
surfaces plus a Mini App, deployed to AWS.
• Took a chat-and-spreadsheet operation to a ~100k-LOC production platform (683 modules) on FastAPI + aiogram +
asyncpg + Redis with Dishka dependency injection. Six products on one shared core: outbound calling operations, a
candidate pipeline with LLM normalization, a laboratory workflow, a doctor consultation board, paid-ad lead intake, and an
internal AI assistant — plus a 22k-LOC shared library layer. Feature auto-discovery means a new product ships without
touching the composition root.
• Built the LLM automation layer. Claude runs as a pool of managed subprocesses with concurrency caps, Redis-backed
session state and classified error handling with user-facing fallbacks; Google Gemini Vision on Vertex AI reads lab forms,
passports and ultrasound reports straight out of Cloud Storage, with an automatic Claude fallback. Free-text and photo input
from staff becomes structured database records with no manual data entry.
• Built the marketing-to-revenue automation, and I also run the marketing it serves. Paid-ad leads arrive by webhook, are
deduplicated, enriched, normalized and routed automatically to the least loaded manager; the funnel is then tracked stage by
stage to the signed contract, with attribution back to the campaign. I spend ~$6.5k a month on Meta and Google myself, so
the system was specified by its own user — the rarest thing a marketing-automation engineer can bring is knowing what the
marketer actually needs.
• Engineered reliable data workflows: an append-only event store, 59 versioned SQL migrations applied transactionally at
boot under an advisory lock, idempotent multi-step pipelines that persist progress and resume instead of duplicating, and
retry-with-backoff on every external boundary.
 • Integrated third-party systems: Telegram Bot API (including a self-hosted local Bot API server to lift the 20 MB file-download
cap), an MTProto userbot daemon for operations a bot token cannot perform, a telephony provider's call API and webhooks,
Google Cloud Storage, and MCP tool servers exposing internal tools to AI agents.
• Run production end to end, alone. GitHub Actions builds a multi-stage image (uv, BuildKit cache mounts, non-root) →
container registry → SSH deploy to AWS EC2 with a health-gated rollout that fails the pipeline if the container is not healthy
within 60 s. The Compose stack is hardened: read-only root filesystems, dropped capabilities, no-new-privileges, per-service
memory limits, healthcheck-ordered startup, an internal-only network for Postgres and Redis, Caddy with automatic TLS, and
a nightly backup cron with a restore-verification script.
• Made quality mechanical, not manual: 6,000+ automated tests across 1,223 modules (~189k LOC) running in ~30 s via
pytest-xdist with a per-worker Postgres clone; a purpose-built mock Telegram server that replays real user flows end to end; a
mypy-clean production surface and ruff/Biome gates in CI. Every feature carries a living technical spec — 96 documents,
~16k lines, kept in sync with the code by a gate, so the system stays legible to anyone who joins it.
• Built an autonomous self-repair engine (~20k LOC): a scheduled pipeline that detects stale, contradictory or unenforced
internal instructions, dispatches LLM fixers in isolated git worktrees, and has every fix judged by an independent zero-context
verifier model — with circuit breakers, planted decoy tasks to detect a broken verifier, revert-on-failure and a full Postgres
audit journal.
• Built the client-facing Mini App — a bilingual form application in vanilla ES2022 modules under strict type-checking, with
native-iOS drill-down navigation and swipe gestures, multi-phone management, client-side validation mirrored by server-side
rules, HMAC-validated Telegram auth and a rate-limited error-reporting endpoint. ~2.2k LOC of JS/CSS under 25 test suites.
Head of Asian Department — client acquisition & marketing 2016 – Present
Own the Asian market end to end: lead generation, advertising, client communication and the team that runs it — held in parallel
with the engineering role above.
• Run client acquisition and advertising for the region — campaign planning and budgets, lead qualification, and the full
communication funnel from first contact to signed client.
• Lead the departmental team and its day-to-day operations. The manual workflows I built and ran here are the ones I replaced
with the automation platform above — I automate marketing operations I am personally accountable for, and I use the
result every day.
• Own the acquisition channel mix — paid-ad lead capture, outbound calling operations, and the qualification funnel that
routes every lead to the right manager. These are the same three surfaces I later rebuilt as automated products.

Enlil, Inc. — MedTech development platform (Campbell, CA) 2022 – 2023

Backend Developer in parallel with the role above
Backend engineering on a regulated medical-device lifecycle platform — design controls, requirements traceability and quality
management for FDA 21 CFR Part 820 / Part 11 and ISO 13485 workflows.
• Built and maintained backend services and REST APIs for the platform's product-development modules, working in Python
against a relational data model with strict referential integrity.
• Worked inside regulated-software constraints: append-only audit trails, versioned records and traceability from requirement
through verification — every data change attributable and reconstructable.
• Collaborated with a distributed US/Ukraine team in English, through code review and written specs.

WOG — West Oil Group, one of Ukraine's largest fuel retail networks 2014 – 2016

Sales Manager — petroleum products
• Bought and sold LPG and diesel: supplier negotiations abroad, rail import, transfer to road tankers, route planning and
delivery — an early lesson in operations that either arrive on time or cost money.

SELECTED PROJECTS

Autonomous self-repair engine — the safety design: planted decoy tasks detect a broken verifier, a circuit breaker closes the
system on three quality failures in a rolling week, environment deaths never consume an attempt, and every limit on cost, wall-
clock, concurrency and attempts is frozen in code and pinned by tests. The models only propose diffs and verdicts.

Vision intake conveyor — Gemini Vision on Vertex AI reading medical documents directly from Cloud Storage with a Claude
fallback, structured-output validation, and per-surface cost and rate-limit budgeting.

Real-time call-outcome classification (R&D, validated in production) — a server-side listener joins live calls and Gemini
classifies what it hears so the system sets the disposition itself. Measured on real traffic: 62 % capture, 4 simultaneous
conversations on one line, ~$0.0015 per classification.

EDUCATION & LANGUAGES

Education Bachelor's degree — Dnipropetrovsk Humanities University, 2010 – 2014
Languages Russian — native · Ukrainian — conversational · English — B1 (Intermediate)