EN RU VI
Ruslan Badaeff

Ruslan Badaev

AI Automation Engineer · GenAI Engineer · Developer-first

Vietnam (since 2022) · Remote · Open to relocation

Email: dev.badaeff@gmail.com · GitHub: github.com/ruslanbadaev

Positioning

I am an AI Automation engineer with a strong software engineering core. I work in a business-first model: validate the hypothesis with a rapid prototype first (n8n/Make/Zapier + scripts), then bring it to production with full integrations, metrics, tests, alerting, and operations. I do not build for the sake of building: the target is always faster execution, fewer errors, higher revenue, and predictable delivery under tight deadlines.

In short: I can launch an MVP quickly, then turn it into a stable 24/7 system without slowing business momentum.

Business Value

Key Domains

TravelTechHospitalityPropTechFinTechIoT/TelematicsSupport automationMarketing automationLead generationTelegram ecosystemsOps toolingRPAWhite-label products

Experience · Case Studies

AI Tour Operator for Automated Tour Assembly

2025 · TravelTech · Vietnam market

Summary: instead of manually assembling one tour in ~4 hours, the system started generating 3 complete tour options from scratch in a single cycle.

  • Context: a Vietnam-based travel company wanted to remove manual routine, pricing mistakes, and dependency on business hours/time zones.
  • Goal: speed up tour production, reduce financial and logistics errors, and provide 24/7 request processing.
  • Implementation: the pipeline accepted incoming requests from clients/partners, generated clarification questions (budget, dates, flights), built the itinerary skeleton around the destination, then attached transfer, hotel, restaurant, and place-of-interest services with seasonality, opening hours, and expected visitor density taken into account.
  • Additional automation: transfer timing checks between points, financial calculation control, alternatives for accommodation and routes, and end-to-end program structure generation for managers without manual recalculation.
  • Before → After: previously a manager assembled 1 tour manually in ~4 hours and the client accepted/rejected a single option; now the system returns 3 curated packages (Budget / Balance / VIP) immediately, giving the client choice within one request.
  • Result: according to client data, conversion increased by 25% due to immediate three-scenario choice and drastic preparation-time reduction. The key effect was not repackaging the price, but moving from one manual option to automated multi-option assembly.
  • Engineering note: started with an n8n prototype, then strengthened with Python/Node.js logic, Flutter Desktop admin panel, server-side production setup, and PostgreSQL storage.

Key metrics: tour generation time, pricing error rate, transfer timing error rate, conversion by package, after-hours coverage.

n8nPythonNode.jsFlutter DesktopPostgreSQLFirebaseLinuxpm2Google Distance Matrix APIGoogle APIsOpenAI APITravel automation

AI Support Layer for Short-Term Rentals on Top of Hostfully

2024 · Hospitality/PropTech · US rentals

Summary: before implementation there was no full after-hours emergency flow; after launch, the system created maintenance dispatches automatically for confirmed incidents at any time of day.

  • Context: the team managed US properties (Airbnb/Booking), and emergency requests outside business hours were waiting until 8:00 AM.
  • Goal: run 24/7 triage and escalation, auto-link each message to the right property_uid/lead_uid/thread_uid, and reduce compensations and negative reviews.
  • Implementation: webhook flow over Hostfully Inbox, emergency-case classification, photo request from the guest, visual validation via Gemini, then automatic escalation by property rules through push/SMS/WhatsApp/email.
  • What happened in the flow: the system immediately identified property/booking/thread, separated routine issues from emergencies, collected photo evidence, assigned a status (not emergency / needs human review / confirmed emergency), and triggered escalation without waiting for the morning shift.
  • Before → After: at night, critical issues could wait until morning or guests had to panic-call for help; now the system accepts requests instantly, runs triage, confirms incidents, and dispatches maintenance when truly required.
  • Result: after-hours first response coverage grew from 0% to 100%, cases waiting until 8:00 AM dropped by 89%, false maintenance dispatches decreased by 38%, average emergency-case resolution time dropped by 61%, compensation rate for emergency requests decreased by 31%, and negative reviews caused by delayed support dropped by 27%.
  • Accuracy and control: property/booking resolution accuracy reached 99.3%, emergency classification 93%, and critical cases triaged before business hours increased from 6% to 91%.

Key metrics: first response time, time to emergency triage, time to maintenance dispatch, false positive dispatch rate, compensation rate per 100 bookings.

Hostfully APIWebhooksNode.jsPythonGoogle Gemini APIPostgreSQLRedisFirebase/FCMFlutter AdminDockerNginxLinuxpm2Incident automation

AI Telegram Infrastructure for Lead Generation and Community Monetization

2025 · LeadGen · Expat communities

Summary: the project moved from random ad posting in groups to personalized outreach exactly when users signaled active demand.

  • Context: before implementation there was no lead-gen system. Ads were posted manually and mostly randomly across groups, without precise matching to active intent.
  • Goal: automate demand-signal discovery, classify message streams cheaply in real time, reduce manual moderation, and improve sales conversion.
  • Implementation: a single classification entry point for spam, intent, commercial signals, and user profiling. Cheap nano models processed most traffic, while stronger models were invoked only for complex context and personalized outreach.
  • What the system captured: open user attributes (geo, visa status, contacts, interests, profession), explicit needs, and commercial intent signals for follow-up touches and segmentation.
  • Before → After: random mass advertising with no readiness signal became personalized communication and ad placement at confirmed moments of user interest/need.
  • Result: signal-to-outreach conversion increased from 7% to 34%, outreach-to-sale +46%, hot-lead loss -81%, cost per 1000 messages -88%, fully automated classification reached 96%, admin workload -67%, and spam moderation speed improved 14x.
  • Scaling effect: on top of the same stack, the client launched AI-targeted ads and increased ad-placement revenue by 185%.

Key metrics: total messages processed, nano-model cost per 1k, lead signal detection rate, outreach-to-sale conversion, advertiser audience match rate.

PyrogramPythonGPT-4.1-nanoGPT-5-nanoFastAPIMongoDBSQLiteRedisFlutter WebWebSocketTelegram Bot APIZapierOpenSearchDockerNginx

RPM: Predictive Monitoring and Support Automation

2019-2020 · IoT/Telematics · InTerra IoT

Summary: moved from reactive support to proactive operations: the team often contacted clients before a visible incident occurred.

  • Context: before implementation, clients noticed sensor failures first, called support, a ticket was created, and only then an engineer was dispatched.
  • Goal: detect pre-failure signals early, automate ticket creation, reduce downtime, and lower negative-review share.
  • Implementation: started with an ML task for fuel-theft detection (separating real theft from false scenarios caused by noise, temperature, and slope), then expanded to telemetry anomaly detection and automatic support-flow initiation.
  • Before → After: client reports after breakdown became system-generated early signal, followed by proactive support contact and pre-failure dispatch planning.
  • Result: operationally, downtime and negative reviews were nearly halved, some repairs became scheduled before critical failure, and the share of incidents first reported by clients decreased.
  • Extension: added object-management operations, role-based permissions, and approval flows for high-risk actions.
  • Public project context: dzen.ru/a/Xs0QNgdbOASNMHoE

Key metrics: predicted failure rate, proactive ticket creation rate, downtime reduction, client-reported incidents share, fuel theft detection precision.

PythonNode.jsJavaMongoDBRedisDockerReactTCP/HTTP/MQTTWialonML modelsStreaming + BatchRPAApproval flows

Automation Layer for a White-Label CORE Product in a Flutter Department

2023 · FinTech/Product Delivery · Head of Department

Summary: transformed team overload from repetitive routine into an internal automation layer that accelerated delivery and improved release predictability.

  • Context: a large part of team time was spent on repetitive processes: docs, reviews, tests, localization, builds, releases, and environment support.
  • Goal: reduce manual routine and speed up white-label CORE delivery without quality loss.
  • Implementation: automated documentation from git diff, auto-generated tests for changed blocks, CI/CD with webhooks and dev/stage/prod separation, plus localization and JSON key automation.
  • Before → After: a heavy manual operational layer around delivery became a systematic pipeline where a significant share of repeatable actions ran automatically and predictably.
  • Result: test coverage grew from 0% to 10%, manual localization/key errors decreased, delivery became faster and more stable, and release flow became more transparent for both team and client.
  • Practical effect: engineers and leads spent less time on operational routine and more on product work.

Key metrics: test coverage growth, release pipeline duration, missed localization keys rate, deployment failure rate, developer routine time saved.

FlutterDartPythonNode.jsGitCI/CDWebhooksLLM APIsJSON localizationLinuxDev/Stage/ProdRelease engineering

Selected Products and Links

Core Skills

AI Automation and GenAI

  • Agentic workflows and orchestration.
  • Multi-step prompting, structured outputs, tool calling, and validation loops.
  • Model routing, budget guards, cost monitoring, and fallback strategies.
  • Production reliability: retries, idempotency, deduplication, runbooks.

Backend and Integrations

  • Python, Node.js, Java.
  • REST APIs, webhooks, queues, event-driven scenarios.
  • PostgreSQL, MongoDB, Redis, Docker, Linux.
  • Integrations with CRM, property-management, Telegram, payment, and internal systems.

Product and Interfaces

  • Flutter (mobile/web/desktop), rapid admin panels, and internal tools.
  • Embedding an AI layer into existing products without full architectural rewrites.
  • CI/CD, release engineering, dev/stage/prod environments.

Delivery Management

  • Short hypothesis cycles and fast production launches.
  • Technical leadership: code review, mentoring, hiring, incident response.
  • Execution under tight deadlines with focus on business outcomes.

Education

MTUCI (Moscow Technical University of Communications and Informatics) - Bachelor - Software Engineering and Intelligent Systems

2018 - 2022

FAQ

Approach in short: value first, then scaling, quality control, and stable operations.

Are you more low-code or code-first?

I am outcome-first. For rapid starts, I use low-code/no-code to validate hypotheses as cheaply and quickly as possible. As soon as business value is proven or platform limits appear, I move critical parts into code, then add testing, metrics, and operational hardening.

How do you ensure AI automation reliability?

I design for failure from day one: timeouts, retries, deduplication, fallback, logging, alerts, and manual takeover. This is essential in support, fintech, and money-sensitive operations where errors are costly.

How fast can we get a working result?

Usually, first production-usable scenarios appear within days. Then we iterate: improve accuracy, broaden case coverage, add team UI, monitoring, tests, and operational integration.

Do you only build bots?

No. I build practical AI layers of all kinds: support automation, lead generation, RPA, documentation and release pipelines, internal tools, interfaces, and integrations with existing products.

How do you calculate LLM and infrastructure economics?

I calculate cost at scenario level, not only per model call. I track cost per resolved case: processing cost, share of manual takeover, avoided errors, and avoided compensations. Usually I use cheap models for high-volume filtering and stronger models only for complex stages.

How do you execute under very tight deadlines?

I split delivery into short iterations with early launch in a limited scope. First, we make the critical business flow work end-to-end; then we add accuracy, coverage, and operational improvements. This gives visible results quickly without getting stuck in perfectionism before first release.

What if data is messy and incomplete?

I include normalization, confidence scoring, and explicit data-quality statuses. If confidence is low, the flow moves to semi-automated mode with prepared context for a human operator. This is better than aggressive full automation on low-quality inputs.

Can you integrate into an existing team instead of rebuilding from scratch?

Yes. Often it is faster and cheaper to embed an AI layer into the client's current processes, APIs, and interfaces than to rewrite the whole product. I work both as a hands-on engineer and as an integration tech lead, from architecture to concrete production fixes.

Which risks do you mitigate first?

Risks that hit money and reputation first: misclassification of critical cases, lead loss, duplicate operations, invalid calculations, missing escalations, and monitoring blind spots. That is why early versions always include metrics, alerts, and manual override.

Video: Automated Content Generation