Development of CV Analyzer and Text2SQL AI Agents

Engineered AI agents leveraging LangChain and Langflow to automate data processing and hiring workflows.

TL;DR: Developed and deployed RAG-based and multi-agent AI solutions at Orbislabs to automate CV screening for Human Resources and translate natural language into SQL for enterprise databases.

OrbisFlow Agent Architecture

Technologies & Methodology

  • Software & Frameworks: Python, PyMuPDF, FAISS vector database, LangChain, OrbisFlow (modified Langflow), and Vanna AI.
  • Models: SBERT, Cohere Embedding, Cohere Rerank, Cohere LLM, and local LLMs.
  • Databases: PostgreSQL.

Key Contributions & Results

  • Developed an RAG-based CV Analyzer that utilizes PyMuPDF for text extraction, Cohere embeddings, and an FAISS vector database to automatically rank candidates against job descriptions.
  • Integrated a Cohere LLM with a memory component to allow users to interactively chat with the CV database, extracting missing or matching candidate qualities while reducing human hiring biases.
  • Engineered a Text2SQL multi-agent architecture using LangChain for an insurance company, enabling non-technical managers to query a PostgreSQL database using natural language.
  • Implemented a complex LangChain routing system containing table-specific agents, fuzzy filter checks, and query validation layers to generate and refine SQL queries.
  • Evaluated and iterated through multiple architectural frameworks (such as SBERT and Vanna AI) and applied prompt engineering to resolve LLM output format errors and reduce hallucination issues.