Naoures boufaied boufaied

Ijoined Venari Security as Senior data scientist and then was promoted to team lead and innovation manager. During this experience, I contributed to several projects, including OS detection and browser identification based on TLS traffic analysis at the TCP/IP layers. Both projects achieved over 98% precision and were integrated in the V-detect tool.
* In recent months, our focus has shifted towards integrating Generative AI (GenAI) to enhance product capabilities. We conducted a feasibility study to explore solutions for new functionalities in the V-Comply platform, targeting areas such as TLS visibility, compliance, hygiene, risk, and vulnerability management.
* we identified several functional use cases for integrating LLMs, including conversational AI and information retrieval, contextualization and summarization, pattern recognition, and TLS traffic insights generation.
*we defined the technical requirements essential to the experiment: cybersecurity expertise, regulatory knowledge, and personalization capabilities.
* we adopted a Retrieval-Augmented Generation (RAG) appr oach, enabling us to enhance LLM outputs by integrating external data (tls compliance regulations) for domain-specific, context-rich responses.
* A first version (MVP) is currently being deployed so that we evaluate the performance on production data in terms of relevance of responses, latency and scalability of the prototype

* During this experience I contributed to developing a solution to automatically assess client solvability, creditworthiness, and risk profile to support banks in making informed loan approval decisions.
* The solution included several key modules: document classification using CNN-based image classification (we achieved 97% accuracy), (OCR for extracting relevant information from documents (we were able to fully extract the information from english/ french written documents and achieved 72% accuracy on arabic documents
* worked on a contextual analysis module that generated a final credit score for each client. Scores were generated with 96% confidence interval.

Developing an NLP toolbox to analyze negative news as part of the client’s screening review module . the goal was to provide the necessary tools for the bank to have a 360 degree view of the client and ensure they are not involved in any criminal activities.
* Data pre-processing techniques to clean/normalize textual data. For example: stop word removal, stemming, light stemming, n-grams extraction, …
* Automation multiclass negative news classification, to be able to automatically classify whether the news article belongs to the criminal class(financial criminality, terrorism, extorsion etc, )
* Leveraging transfer learning, we were able to achieve a 98% f1-score using only 100 labels
* Named entity recognition and search similarities relative to customers personal information.
* Deployed the pipeline within the KYC/ screening review module
in the CENTAURE solution, completely automating and reducing the time