Cypress.io

Senior Software Engineer

FindHotel

Software Engineer

EGT Ukraine

- Worked on creating gambling/betting b2b platform;
- Redesign system data consumption of provider data to separate feed into two parts;
- Implemented initial integration e2e tests and set up CI for the system;
- Created API for websocket event subscription;
- Implemented cache purging for Cloudflare for particular events from Kafka;
Tech: Go, SOA, grpc, Kafka, Postgres, Kubernetes, gitlab CI.

- Created web crawling and scraping system to parse craigslist cars/tracks as fast as possible. 
Tech: Go, AWS, colly, MySQL, telegram-bot-api.

- Created service for crawling proxies and providing gateway to rotate them (as SaaS).
Tech: Go, AWS, microservices, grpc, MongoDB.

- Created different bots for automation routine tasks, like monitoring and gathering data into excel report.
Tech: Go, Heroku, telegram-bot-api, excel.

- Created pubmed scraper collect to parse articles.
Tech: Node.js, mongoDB.

- Implemented unit and end-to-end tests for the system for development and set up CI.
Tech:  Go, docker-compose

- Created a service for NLU and NER that could label input text based on provided training data.
Tech: Node.js, nlp.js

- Implemented different API services, like jogging tracking.
Tech: Go, microservices, grpc, Mongo, swagger.

- Built a Terraform script to set up AWS EKS and MSK and add support of the confluent schema registry for the infrastructure.

Freelance

Bitfury

- Worked on creation of Bitcoin payment processing hub that supports lightning payment
- Modified open-source code of Lightning Network Daemon to add functionality that we needed
- Implemented security improvements with ECDSA signatures
- Implemented different microservices, like email, firebase, user notifications via web sockets and other
- Made improvements to the deployment process
Tech: Go, microservices, grpc, Circle CI, MongoDB, bitcoin, lightning, AWS, Nomad, Vault, RabbitMQ

- Managed team of 2 frontend developers
- Developed from scratch bitcoin open-source wallet with lightning payment support for desktop and created CI and CD of the wallet
- Created backend for the wallet with Python, Django and PostgreSQL
- Participated in different researches where we analyzed our different blockchain use case ideas like scaling other blockchains transactions per second capacity
- Collaborated with the head of project, designers and QA engineers to make a high-quality and user-friendly product
- Made prototypes for our internal ideas of platforms
- Created payment plugins for Magento 2 and Woocommerce to support lightning payments
Tech: Node.js, React, electronjs, bitcoin, lightning, Go, Nomad, open-source, research.

Team Lead of front-end team

- Made algorithms to solve customers problems, like solving custom puzzles that were invented by client
- Created web applications for customers
- Created scripts and tasks for a trading platform with mql4
Tech: React, jQuery, javascript, MQL4, pascal.

Created some strategies and implemented them using Metatrader 4 and MQL4 based on neural networks and statistical analysis.

Analytics

Analytical, quick learner, resilient, self-starter, intelligent, result-oriented

90% of companies using "AI" are actually just using fancy autocomplete. 

Here's the problem: 

They're stuck in Level 1 thinking - asking AI questions and manually implementing everything themselves. 

But there are actually 4 distinct AI power levels: 

🔹 LLMs - Text prediction models 
🔹 Automations - Scripted business workflows 
🔹 Agents - Systems that reason and decide autonomously 
🔹 Swarms - Multiple agents working as coordinated teams 

The real breakthrough happens at Level 3. 

Unlike automations that break when something unexpected occurs, agents can handle situations you never explicitly programmed. 

Think: An AI that manages client onboarding, adapts to different document formats, predicts which clients need extra attention, and solves problems in real-time. 

That's not automation. That's autonomous agent. 

Most companies will stay stuck at Level 1. 
The ones that don't will dominate their industries. 

What's keeping your organization from moving beyond basic AI tools? 

#ArtificialIntelligence #AI #AIStrategy #gpt

[Watch the full breakdown - link in comments]

Vlad Budichenko

AI voice agents for businesses 📞 | Founder @ Vocaly AI | Software engineer & Solution Architect 💻

🎬 Hooked up VEO 3 to my n8n setup. 

Now it just makes viral videos on its own.

Here's what the workflow does:

💡 Step 1: Ideation agent brainstorms concepts ↓ 
✍️ Step 2: Prompt agent writes VEO 3 instructions ↓
🎬 Step 3: Video generates via Replicate API ↓ 
📁 Step 4: Auto-download & spreadsheet storage

Built with OpenAI chat models handling the reasoning, n8n managing the workflow, and VEO 3 through Replicate doing the heavy lifting.

Perfect for people who want to automate their content pipeline. 🚀

The whole thing runs without me touching anything after the initial setup.

Like + comment "WORKFLOW" & I'll send you the complete n8n setup (must be connected so I can DM)

#n8n #VEO3 #Automation #AIWorkflow #VideoGeneration

New video about Social media automation with n8n and make.com

#ai #automation #n8n #makecom

The question every voice AI startup will hear next week: "How are you different from OpenAI's API?". Here's how 1,000 founders should answer. 🎯

OpenAI's Real-Time API just became the new baseline.

A little bit better function calling? Sure. 🤷 Slightly faster responses? Great.

But let's be honest - this is incremental improvement, not the breakthrough everyone expected (like in GPT-5 😉).

The good news? The real opportunities were never in general conversation anyway.

Here's how to survive the "Why not just use Open AI?" question:

→ Domain Expertise 🏥 
"We fine-tune on medical/legal/technical vocabulary that general models can't handle reliably." 
→ Data Control 🔒 
"We train on proprietary datasets and deploy offline for compliance-sensitive industries." 
→ Multi-Model Orchestration 🧩 
"We combine specialized models - one for transcription, one for reasoning, one for generation - optimized for our specific workflow." 
→ Custom Conversation Patterns 💬
"We've built conversation flows for [specific use case] that generic APIs weren't designed for."

The startups that survive won't be building "better ChatGPT." 

They'll be building specialized intelligence for problems OpenAI's one-size-fits-all approach can't solve. 🎪 

Because when everyone has access to the same foundation model magic, differentiation happens in: 
→ The specific problem you solve 
→ The unique data you train on 
→ The specialized workflow you optimize for 
→ The compliance requirements you meet 

OpenAI just commoditized general voice conversation. The opportunity is in everything they didn't optimize for. 

💡 What specialized voice AI application do you think breaks through first? 

#VoiceAI #OpenAI #TechReality #AI

🐷🚫 Spam callbacks and inbound calls are quietly draining voice-AI budgets: fake forms trigger expensive callbacks; robodialers or prerecorded spam eat minutes fast.

The fix is a small stack of gates that auto-tighten when abuse is suspected—so real clients aren’t hit with random checks.

📞 Callback abuse (form → auto-dial)

🔹 Gate the trigger: invisible CAPTCHA + honeypot + ≥3–5s dwell; require a short-lived server signature (≤60s) on any call job. No signature → no call.
🔹 Throttle by number: allow natural retries, then cool-down (e.g., up to 2 retries within ~20 minutes after no-answer/busy; then 6–12h cool-down; collapse duplicates within 30 min).
🔹 Reputation first:  number lookup for spam score could be helpful (like most can add Twilio Nomorobo add-on in case of such provider, $0.003 USD per request).
🔹 Graylist verification: in case of prefer low-friction checks - email-link confirm or one-ring “flash” call; if IVR DTMF is used, keep it 2–5s max. (Typically fractions of a cent per attempt; we often skip it for UX/cost.). Manual review helps if necessary.
🔹 Cost guardrails (like this one 💰): per-source daily cap (e.g., $50/day), per-number cap (e.g., $2/day), and a circuit breaker works great, that can pause auto-callbacks, route to graylist/voicemail, alert, then auto-relax after cooldown.

📲 Inbound spam calls
 
🔹 Trust signals can help: Spam score or use STIR/SHAKEN; downrank “None” or mismatched CNAM and route low-trust to screening. (Same sub-cent lookup.)
🔹 First-seconds triage: transcribe the opening 5–10 seconds; flag phrases like “Google listing,” “urgent verification,” “final notice.” Low trust + phrase hit → voicemail with transcript or drop. (Short snippet ≈ sub-cent.)
🔹 Evolving lists: auto-promote verified callers to allowlist (7–30 days); auto-denylist rapid repeats/failed screens, with score decay so false positives recover.
🔹 Minimal friction when needed: if you dislike IVR DTMF (like me 🤮), rely on silent screening + allowlist; only enable a 2-digit challenge during spikes. (Still ~ fractions of a cent.)
🔹 Anomaly controls: per-campaign concurrency/hourly/daily spend caps; on a 3× spike or rising invalid rate, auto-elevate screening globally, then relax when metrics normalize.

Run “adaptive friction” to not annoy real clients. In normal traffic, trusted/repeat numbers bypass screens; on abuse signals the system auto-enables graylist/screening/caps; when clean again, it dials back. This keeps verification costs low and protects CX.

If you’re scaling voice AI and want to explore practical ways to implement this stack in your flow, DM me 📥 happy to walk through options. 

#VoiceAI #AI #BotProtection #Bot #Spam

Taplio

Vladyslav Budichenko 🇺🇦's Linkedin Analytics

's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Vladyslav Budichenko 🇺🇦's Linkedin Analytics

's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Vladyslav Budichenko 🇺🇦's Linkedin Analytics

's Best Posts (last 30 days)