👋 Hello, I'm
Keval Savaliya
Agentic AI Engineer & ML Specialist building intelligent systems, from MCP tool integrations and multi-agent pipelines to production-grade ML models and RAG-powered applications.
Agentic AI Engineer & ML Specialist building intelligent systems, from MCP tool integrations and multi-agent pipelines to production-grade ML models and RAG-powered applications.
I'm an AI Engineer focused on Agentic AI and AI Automation, currently pursuing my Master's in Automotive Software Engineering at TU Chemnitz, Germany. I'm also working as an AI/ML Research Intern at Mindioo, building recommendation systems and LLM-powered features for an AI marketplace platform.
With 1.5+ years of hands-on industry experience, I specialize in multi-agent architectures, RAG pipelines, MCP integrations, and n8n automation workflows.
Seeking a working student position or Master's thesis to apply AI engineering expertise to real engineering challenges. Immediately available, student visa authorised for part-time work in Germany.
Ingested a medical PDF corpus, split into chunks, embedded with SentenceTransformers (384-dim), and indexed into Pinecone serverless on AWS — entire ingestion runs as a single script. Added conversation memory so follow-up questions are rewritten in context before hitting the vector index. All answers grounded via Google Gemini — zero hallucinations on clinical queries. Containerised with Docker and GitHub Actions pushes to AWS on every merge.
Built a RAG system in LangChain that ingests YouTube video transcripts, embeds them into a vector store, and enables conversational Q&A over video content. Demonstrates the full retrieval-augmented generation loop using LangChain's document loaders, text splitters, and chain abstractions.
Two production MCP servers exposing 7 AI tools directly integrable with Claude and any MCP-compatible client. Includes a TypeScript MCP client connecting to the Python server for cross-language agentic integration, combined into a Claude Skill for clinical software developers.
End-to-end RAG pipeline for e-commerce customer support: PDF policy ingestion, text chunking, semantic embedding with SentenceTransformers, and vector storage in ChromaDB. Agent performs semantic search to retrieve relevant policy chunks and generates grounded answers - achieving zero hallucination on policy queries.
End-to-end ML system forecasting PM10 air quality levels using XGBoost, achieving R² of 0.65 and 24% improvement over baseline. Production REST API with automated daily data pipeline from OpenAQ, interactive Streamlit dashboard, and a RAG-based health assistant providing personalised WHO/EU guideline advice across 6 health profiles.
NLP classification system using fine-tuned BERT transformers achieving 81.5% F1 score on 7,613 disaster tweets. Applies semantic analysis to distinguish real emergencies from metaphorical disaster language. Deployed as an interactive Streamlit interface with real-time predictions and model performance visualisations.
Intelligent email monitoring system using LLM-based content analysis to automatically detect interview invitations, achieving 90%+ accuracy with automated data pipeline integration.
Interactive AI application leveraging Claude Sonnet 4.5 for personalized workout recommendations and real-time guidance with dynamic workout adjustments.
Comprehensive analytics dashboard monitoring KPIs and consumer behavior, identifying 20%+ revenue optimization opportunities with extensive SQL data quality audits.
Time-series forecasting system using Stacked LSTM neural networks, achieving 85% prediction accuracy with real-time data acquisition from yfinance API.