Djordje Vuckovic

• • Designed and built "Advanta", a production-grade financial Al chatbot that analyzes structured and unstructured datasets to detect customer churn rates and data quality issues for banking clients.

• •Architected an Al agent-based system with Retrieval-Augmented Generation (RAG) components, enabling natural language queries, advanced analytics, and interactive data visualizations.

• • Integrated Claude 3.5 APIs for reasoning, insight generation, and contextual responses, combined with VVDB vector embeddings for semantic search across diverse datasets.

• • Developed backend services in Django with a React frontend, implementing a JSON-based visualization protocol for generating graphs, charts, and geographic maps dynamically.

• • Optimized system performance through indexing, caching, and asynchronous processing, achieving sub-second retrieval speeds for large datasets.

• • Implemented PostgreSQL for user management, access control, and persistent storage of structured data; leveraged AWS cloud infrastructure for deployment, scalability, and reliability.

• • Dockerized the full application for portable, environment-agnostic deployments across development, staging, and production environments.

• • Established data ingestion pipelines for both tabular and unstructured content, enabling automated feature extraction and analysis.

Built an Al-powered employee onboarding platform combining advanced LLM workflows with robust, scalable full-stack architecture.

• • Developed a real-time RAG pipeline using LangChain and Pinecone with hybrid keyword/vector search across 50+ internal documents, optimized with Redis caching to reduce API costs by 40%.

• • Designed and deployed a conversational chatbot using Claude-3.5 and GPT-4 (via Together.al) with dynamic prompt engineering, managed via WebSockets in Node.js and TypeScript.

• • Built and maintained a modular FastAPI backend in Python for managing onboarding logic, workflows, and chat context persistence.

• • Engineered a responsive Al avatar interface in Next.js and React, integrating real-time audio/video streaming, ElevenLabs TTS, and secure RDBMS data access for user-specific rendering.

• • Integrated Redis Pub/Sub with AWS Glue to trigger cost-efficient workflows, reducing HeyGen API usage by 30%.

• • Led CI/CD infrastructure using GitLab pipelines, Docker, and mypy pre-commit hooks, improving deployment consistency and reducing PR review cycles by 40%.

• • Delivered production-grade Ul with Tailwind CSS, state management via Zustand, and micro-frontend architecture for modular onboarding components.

• • Developed internal scraping tools to extract structured content from company handbooks, policy documents, and onboarding assets, enabling dynamic RAG indexing across 50+ sources with Claude-3.5 and Pinecone.

• •Applied custom text cleaning, PDF/HTML parsing, and chunking logic to ensure high-quality embeddings and context-aware retrieval accuracy.

• • Built an Al-driven semantic chatbot using LangChain and Pinecone, integrated with domain-specific NLP pipelines for real-time responses.

• • Built robust, large-scale data scraping pipelines for WDI, Eurostat, and IMF APIs, ingesting 5M+ records monthly with historical backfills and daily deltas to power semantic search and Al-driven insights.

• •Applied anti-blocking tactics including IP rotation, user-agent spoofing, conditional requests, and backoff retries, ensuring 99.9% ingestion reliability in production.

• • Designed and maintained robust backend APIs using NestlS and FastAPI, handling secure data ingestion, transformation, and storage.

• • Implemented a React + TypeScript frontend with dynamic charts and responsive dashboards to visualize real-time economic metrics.

• • Ensured high system performance through asynchronous communication between NLP services, APIs, and vector stores with Redis caching.

• • Built CI/CD pipelines using GitLab and Docker, improving deployment speed and minimizing service downtime.