Yacin Christian Baltagi

- Document Anonymization Pipeline On-Premise

Designed and deployed an on-premise, AI-powered anonymization service for sensitive enterprise documents — combining NLP-based PII detection on text and a locally hosted Vision Language Model (VLM) for image content analysis under strict GDPR constraints with zero data exfiltration.

- RAG Infrastructure

Architected a generic, reusable RAG service as the shared document-querying foundation for all AI Factory squads, owning the full ingestion pipeline from raw document processing to hybrid retrieval.

Stack: Azure AI Search, Azure OpenAI, AKS, Presidio, Ollama, Python, FastAPI, Docker

End-to-end development and architecture of RagMaker, an Agentic B2B SaaS platform enabling users to query multi-format documents and connect external tools via a conversational interface.

🔹 Advanced AI Agents & RAG Core:

• Led a rigorous R&D and benchmarking phase across the entire technical stack, ensuring high retrieval quality and system performance.

• Engineered the RAG pipeline using PostgreSQL (pgvector) to index and query local multi-format files (PDF, DOCX, Images, CSV) with Hybrid Search (semantic, key-word and graph).

• Developed Text-to-SQL capabilities: the agent translates natural language into SQL, executes the query, and renders interactive visualizations.

• Standardized connectivity with external ecosystems (Notion, Drive, Custom APIs) by implementing the MCP (Model Context Protocol).

• Designed a smart agent system with LangGraph capable of autonomously choosing the right tool (Documents, SQL, MCP Tools) to answer complex user queries.

🔹Full-Stack Architecture & Industrialization:

• Built a modern, responsive User Interface using Next.js 15 and React 19.

• Developed a high-performance asynchronous API with FastAPI & SQLAlchemy, optimized for low-latency AI responses and clean architecture.

• Co-designed a scalable, containerized architecture on GCP using Docker and Supabase for managed services.

• Created a Python SDK providing strict typing and clear documentation, simplifying integration for third-party developers.

🛠 Stack:

- LangGraph, RAG (Hybrid Search), LangChain, Embeddings, PostgreSQL (pgvector), Text-to-SQL

- FastAPI (Async), Python, SQLAlchemy 2.0, Hatchet (Distributed Tasks), Supabase, Next.js

- Docker, GCP, SDK Design, MCP

Focused on the development and deployment of Generative AI solutions to streamline internal business operations. Specialized in building secure RAG systems and optimizing LLM inference infrastructure.

🔹 Customer Intelligence RAG Engine (Self-Hosted):

• Evaluated and selected open-source RAG frameworks based on retrieval quality, operational constraints and self-hosting requirements.

• Deployed and integrated the final architecture locally to analyze client feedback from diverse sources (Twitter, Zendesk, engagement/unsubscription forms).

• Engineered BigQuery extraction pipelines with data cleaning and metadata enrichment (Source, Churn status) to enable precise retrieval filtering.

• Deployed a daily automated job to ingest and vectorize fresh feedback, ensuring the Vector Database is continuously synchronized.

• Developed a conversational interface (Streamlit) allowing non-technical teams to extract insights in natural language, significantly reducing time-to-insight.

🔹 LLM Production & Infrastructure (LLMOps):

• Designed and deployed secure FastAPI microservices to expose internal, self-hosted LLM endpoints, replacing legacy ML models.

• Developed robust API contracts using Pydantic for data validation and auto-generated OpenAPI/Swagger documentation.

• Led strategic model evaluation (Open Source vs. Proprietary), benchmarking Cost vs. Performance to assess when LLM-based endpoints could replace or complement legacy ML systems.

🛠 Stack : Python, LangChain, FastAPI, Pydantic, Docker, LLMOps, Generative AI, RAG, BigQuery, PostgreSQL (pgvector), Streamlit, LLM Benchmarking, Data Engineering.