We developed an end-to-end Retrieval-Augmented Generation (RAG) system designed for intelligent document processing and semantic search over large PDF collections. The system is powered by a local fine-tuned LLaMA model and integrates FastAPI for backend orchestration, Qdrant as a high-performance vector database, and Streamlit for an interactive user interface. Key features include PDF ingestion, automatic embedding generation, predictive tagging using KeyBERT, contextual document retrieval, and a modular architecture ready for further AI agent extensions.

	•	Automated document classification and tagging improved discoverability and metadata richness.
	•	Local LLM reduced cost and improved data privacy, especially for proprietary documentation.
	•	System is modular and extendable — future plans include RAG agents and voice-based querying.

Our second branch  DatameshEngine product enables plug-and-play replication between any source and destination database—relational, NoSQL, or cloud-native—supporting peer-to-peer (P2P) replication, blockchain-based synchronization, and high-performance streaming integration via Kafka.


Specialties:

	•	Real-time database replication and synchronization
	•	LLM + RAG systems with local models (LLaMA 3, Mistral)
	•	LoRA fine-tuning and deployment for domain-specific tasks (CVs, contracts, finance)
	•	AI observability: Prometheus + Grafana for LLM applications
	•	Cloud-native and on-prem AI integration (FastAPI, Ollama, Qdrant, Kafka, S3)

Our flagship products:
	1.	DataSync P2P – distributed database replication and synchronization
	2.	RAGBase AI – local RAG system with PDF ingestion and monitoring
	3.	LoRA Forge – tool for training and exporting LoRA adapters (GGUF-ready)
	4.	InsightProbe – observability system for LLM applications
	5.	FastDoc Tagger – CV & document tagging powered by fine-tuned LLMs

DataMeshSync

Org chart

Teams

Offices