Objective

Develop a suite of internal AI tools to streamline recruitment and sales operations, focusing on privacy and cost-efficiency through local LLM deployment.

The Challenge

Standardizing and speeding up manual internal processes (screening CVs, matching profiles to positions, and reformatting documents) without relying on expensive, third-party cloud APIs that pose privacy risks.

The Solution

• CV Screener: Automated tool that screens candidate profiles based on geographical and linguistic criteria.
• Profile-Position Recommender: Matching engine that suggests the best consultants for open roles.
• Style Transformer: Generative tool that maps personal CVs to the standardized in-house template.
• Set up and optimized local LLMs (Llama/Mistral) on AWS for data privacy.

Objective

Develop a multilingual, multi-agency chatbot system capable of handling complex customer queries across different banking sectors.

The Challenge

Managing multilingual support while integrating Retrieval Augmented Generation (RAG) to ensure the chatbot's answers were grounded in proprietary, up-to-date data.

The Solution

• Built a RAG-enhanced search engine and conversational interface.
• Integrated RAG to bridge static LLM knowledge and dynamic internal documentation.
• Enabled Multi-Agency support for a single architecture serving multiple business units.

Objective

Migrate critical ETL processes from a legacy Windows environment to a modern Red Hat Linux infrastructure.

The Challenge

The legacy system relied on ~200 disparate PowerShell scripts—a maintenance nightmare that was prone to failure and difficult to monitor.

The Solution

• Translated and consolidated ~200 PowerShell scripts into <30 Python and Shell scripts.
• Migrated the entire environment to Linux.
• Leveraged Git and VSCode for a version-controlled developer experience.

Objective

Design a system to map professional skills and provide intelligent job/course recommendations using a structured Knowledge Graph.

The Challenge

Skills data is often messy and unstructured. Creating a meaningful recommendation engine required extracting precise entities from vast amounts of text.

The Solution

• Developed a Skills Ontology using Transfer Learning and NER.
• Fine-tuned Transformer models for NER and Entity Linking.
• Used ChatGPT bot to generate synthetic training data with semi-supervised learning.