AI Events concierge (Full-Stack Agentic System)

Event Concierge System Design

End-to-end automation for event discovery, registration, and calendar scheduling. Powered by LangGraph, AutoGen, and browser automation, it acts as a personal concierge—searching, ranking, registering, and scheduling events from platforms like Meetup and Luma. The system orchestrates multiple AI agents and browser automation (Playwright, AutoGen) to handle form filling, event registration, and Google Calendar API integration. State management ensures user info and event status are tracked throughout the workflow, keeping your schedule up to date automatically.

Technologies: Python, LangGraph, AutoGen, Playwright, Google Calendar API, Browser Automation GitHub Repository YouTube Walkthrough

Event ingestion and ranking pipeline (Full-Stack Project)

Event Ingestion System Design

Cloud-native, serverless platform for ingesting, enriching, and ranking event data from diverse sources (e.g., Apify crawlers via webhooks). Built with SST and AWS, it features modular Lambda functions, DynamoDB storage, OpenSearch indexing, and automated publishing to external platforms (Bluesky, Twitter). Supports hybrid search, analytics dashboards, and extensible integrations (OpenAI, Apify). Designed for reliability, scalability, and research-driven workflows, with infrastructure-as-code and CI/CD automation.

Technologies: TypeScript, SST, AWS Lambda, DynamoDB, OpenSearch, Apify, OpenAI, Step Functions, CI/CD GitHub Repository

DSPy: Declarative Language Programs (AI/ML Research)

DSPy System Design

Comprehensive resource for experimenting with DSPy—a framework for declarative, trainable language model pipelines. Includes code, Jupyter notebooks, demos, and documentation for evaluating, orchestrating, and optimizing LLM workflows. DSPy reframes prompt engineering as a machine learning problem, enabling systematic evaluation, reproducibility, and scalable pipeline design. Key features: modular pipeline composition, automated prompt and parameter optimization, and robust evaluation using datasets and metrics. Demos illustrate basic, compiled, and multi-stage DSPy workflows for company and market analysis.

GitHub Repository YouTube Walkthrough

Transformers Fine-Tuning (LLM Research)

Comprehensive exploration of transformer model fine-tuning, focusing on grammar tasks and optimization strategies. Covers theoretical foundations, architectural considerations, and practical techniques for customizing models like T5, GPT-2, and Llama-2. Demonstrates evaluation frameworks, metrics (perplexity), and hardware-efficient methods (PEFT, LoRA, BitsAndBytes, soft prompts). Includes hands-on walkthroughs using Jupyter notebooks and Axolotl, with source code for evaluation, fine-tuning, and infrastructure.

Technologies: Python, PyTorch, Hugging Face Transformers, LoRA, BitsAndBytes, Axolotl, Jupyter, Model Evaluation, Cloud Infrastructure Transformer Research Video Fine Tuning Video GitHub Repository

Cloud Service Providers Blogs Summarizator (Full-Stack Project)

Blog Summarizer System Design

Cloud-native automation for blog summarization and content management, leveraging AWS Step Functions, Lambda, OpenAI, Notion API, and Apify. Crawls, summarizes, and organizes blog content using serverless architecture and LLMs. Features include automated blog summarization (OpenAI, LangChain), scalable content crawling (Apify), secure API endpoints, Next.js frontend, and infrastructure as code (SST, AWS CDK). Extensible to multi-cloud environments and CI/CD ready.

Technologies: Next.js, AWS Lambda, Step Functions, OpenAI, LangChain, Apify, Notion API, SST, AWS CDK GitHub Repository

Personal Website (Full-Stack Project)

Personal Website System Design

Modern web application built with Next.js, serving as the frontend for iliazlobin-sst cloud automation and blog summarization platform. Provides a fast, scalable interface for viewing and managing summarized blog content and automation workflows. Features include modular app directory structure, live reload, font optimization, custom API integration, Vercel deployment, TypeScript, and Tailwind CSS.

Technologies: Next.js, React, TypeScript, Tailwind CSS, Vercel, SST, API integration Demo Video

Twitter Recommendation System (ML Research)

Twitter Recommendation System Design

In-depth analysis of Twitter's open-sourced recommendation algorithm, exploring its system design, technical documentation, and key scientific papers. The video covers how Twitter ranks and recommends tweets, the role of user engagement prediction (RealGraph), feature-wise multiplication in ranking models (MaskNet), community-based representations (SimClusters), and multimodal interaction graphs (TwHIN). Includes walkthroughs of official source code, technical docs, and practical insights for users and content creators. See Twitter's open-source algorithm.

Technologies: Python, Scala, Machine Learning, Graph Algorithms, Recommendation Systems, Twitter API YouTube Video

Voicematch Models: Speech & Audio Analysis Toolkit (ML Research)

Voicematch Models System Design

Toolkit and container suite for advanced speech/audio analysis, including word, phoneme, and pitch evaluation. Provides ready-to-deploy Docker images, model artifacts, and serving scripts for local/cloud deployment (AWS SageMaker, TorchServe). Features custom Python inference handlers for Hugging Face Transformers (Wav2Vec2) and TensorFlow SPICE, flexible configuration, and utilities for local validation.

Technologies: Python, PyTorch, Hugging Face, TensorFlow, Docker, TorchServe, AWS SageMaker GitHub Repository

VoiceMatch: AI-Powered English Pronunciation Practice Platform (Full-Stack Project)

VoiceMatch Demo

Designed and implemented a full-stack, cloud-native application enabling users to improve English pronunciation through advanced speech analysis and interactive feedback. Architected scalable backend APIs, developed modern Vue.js and React frontends, and integrated machine learning models (PyTorch, TensorFlow) for phoneme and pitch detection. Leveraged AWS S3, Docker, and serverless technologies for robust cloud video/audio processing. Delivered real-time, data-driven insights and visualizations, supporting both individual learners and educational institutions.

Technologies: Python, TypeScript, Vue.js, React, AWS, Docker, PyTorch, TensorFlow, Serverless. GitHub Repository

Atmos Landing Zones (IaC Project)

Comprehensive implementation of AWS Landing Zones using Cloud Posse Atmos, Terraform, and Helmfile. Enables secure, scalable, and automated provisioning of multi-account, multi-region AWS environments with modular infrastructure as code. Features include automated account creation, IAM role delegation (SSO/SAML), centralized networking, security guardrails, audit logging, and Kubernetes orchestration. Supports CI/CD automation, reproducible developer environments (Geodesic shell), and extensible configuration for business units and future cloud providers.

Technologies: Atmos, Terraform, Helmfile, AWS, SSO, SAML, Geodesic, CI/CD, Docker GitHub Repository