Pioneering AI and LLM solutions for Advanced Telecom networks

We are a team of dedicated researchers that aim to create Secure, Efficient and Capable LLM solutions.

Signup for Wishlist

About us

Introducing WISIGHTS Lab

We are the creators of ORANSight, a flagship suite of LLMs designed for O-RAN. These open-source models deliver state-of-the-art performance as specification assistants, automated code generators, and comprehensive guides for O-RAN and 3GPP standards. WiLM-Sec (Wireless Language Model Security): We are the creators of the only comprehensive solution that can safeguard wireless or telecom-based LLMs from security threats like prompt injection and jailbreak. WiLM-Sec will also be open-source and updated regularly to meet the security needs of wireless-centric LLMs.

  • ORANSight 2.0, our flagship LLM, functions as a specification assistant, automated code generator, and guide for 0-RAN and 3GPP specifications.
  • WiLM-Sec:Wireless Language Model Security. Our state-of-the-art solution is designed to safeguard language models in wireless environments.
  • We can also build Domain specific LLMs or GenAI solutions for your needs.

O-RAN

0-RAN Alliance on LLMs

The 0 -RAN Alliance is championing the use of Large Language Models (LLMs) to transform open RAN networks. LLMs are set to enhance network efficiency by:

Automating Network Analytics

Identifying patterns and anomalies in data to improve issue detection and resolution.

Generating and Managing Code

Creating IaC scripts. validating code. and maintaining up-to-date configurations for efficient deployment.

Creating Knowledge Graphs

Automating network planning and troubleshooting through comprehensive component relationships.

Generating Documentation

LLMs can automate documentation by extracting key details from network configurations, change logs, and other data sources to create up-to-date records.

These advancements promise to drive greater automation, efficiency, and reliability in 0-RAN networks.

The Existing LLMs do not work for O-RAN

  • To assess the knowledge of existing LLMs for 0 -RAN we leverage the ORAN-Bench-13K benchmark, which consists of 13,952 meticulously curated multiple-choice questions generated from 116 0-RAN specification documents.

  • When we test the popular and widely used LLMs like OpenAI’s Chat-GPT, Google’s Gemini, and an open-source Mistral, the models fail and our initial work named ORANSight provides SOTA performance.

  • In this work, we also introduce the first version of ORANSight, a Retrieval-Augmented Generation-based framework and, notably, the first LLM solution in the literature for O-RAN. The paper has been accepted for publication at IEEE CCNC. https://arxiv.org/abs/2407.06245

Security Challenges of Existing LLMs

  • The available LLM implementations are not secure and are susceptible to multiple prompt-based vulnerabilities.
  • A malicious user can extract and manipulate private and sensitive data through seemingly simple prompts, and commercially available LLMs are not protected against these threats.
  • The security is usually provided through separate vendors which often cause compatibility and licensing issues along with added costs.

An attacker attempts to indirectly prompt LLMs integrated in applications

Injection Methods

  • Passive methods (by retrieval)
  • Active methods (e.g. emails)
  • User-driven injections
  • Hidden injections

Affected Parties

  • End-users
  • Developers
  • Automated systems
  • The LLM itself (availability)

Introducing ORANSight 2.0

Our credentials also include the ORANSight 2.0 release, which comprises a total of 18 models ranging from 1B to 70B parameters and supports a context window of up to 128k tokens. These models are open-source and available here.

With advanced analytics and real-time decision-making, O-RANSight enhances network orchestration and resource optimization. This ensures more efficient, scalable, and reliable operations across O-RAN ecosystems

O-RANSight can generate high-quality code and automate testing processes, significantly reducing development cycles and minimizing human error. This enhances productivity and ensures robust, error-free integrations in the O-RAN ecosystem.

By leveraging the ability to process diverse data types—such as text, telemetry, and images—O-RANSight delivers unified insights. This enables more precise decision-making and better adaptability to complex O-RAN scenarios.

O-RANSight is designed to act as a corrective model capable of observing issues in real-time and dynamically resolving them on the fly. This reduces downtime, enhances reliability, and lowers operational costs.

WiLM-Sec: Wireless Language Model Security

WiLM-Sec continuously monitors user inputs to identify and mitigate security risks specific to wireless LLM environments. This system uses sentinel tokens—specialized markers embedded within the prompt—to detect tampering attempts and assess whether an input might be malicious. If an input contains these sentinel tokens in unexpected or unusual ways, WiLM-Sec initiates additional checks and validation steps to confirm whether the input is safe.

By leveraging deep packet inspection and contextual analysis, WiLM-Sec effectively detects and neutralizes both known and unknown threats in real-time, ensuring that your LLM remains secure and resistant to manipulation in the open, dynamic setting of wireless networks.

To address the unique security challenges in wireless LLM environments, we have built WiLM-Sec with an automated leak detection and reporting system. If a potential data leak is identified, WiLM-Sec immediately logs the event and generates a detailed report. This includes information on the type of injection attempt, the input pattern, and the system response.

This real-time logging ensures transparency and offers insights into any attempted breaches, enabling your team to investigate and address issues quickly. The leak reporting feature maintains a record of incidents, helping to refine WiLM-Sec's security protocols continuously.

WiLM-Sec is designed for straightforward integration with existing wireless infrastructures, supporting popular frameworks like LangChain, Hugging Face Transformers, and OpenAI’s API. This compatibility enables quick and secure deployment across various LLM applications with minimal setup, ensuring your operations run smoothly and efficiently. Whether you’re working with a single LLM model or a suite of interconnected applications, WiLM-Sec allows for seamless scaling and robust security across all deployments.

By leveraging trusted frameworks, WiLM-Sec ensures organizations can confidently secure their LLM-powered solutions without additional complexity or performance overhead.

We continuously update WiLM-Sec to adapt to emerging vulnerabilities, ensuring the platform stays ahead of potential threats. Our commitment to proactive defense includes regular security patches and enhancements based on real-world data from threat analysis and incident logs.

These updates are managed seamlessly, requiring minimal intervention and allowing WiLM-Sec to evolve with the security demands of wireless LLM environments. By staying current with the latest advancements in LLM security, we ensure that WiLM-Sec remains a resilient and future-ready solution for safeguarding your sensitive communications.

Services

GenAI Consulting

Specialized AI services, including large language model development and integration, tailored for telecom and enterprise use cases.

Telecom Consulting

Strategic guidance for optimizing network performance, O-RAN deployment, and future-proofing operations

Applied Research & Development

Custom R&D solutions for cutting-edge telecom technologies and O-RAN advancements.

Meet the team that makes it happen

Meet our diverse team of world-class creators, designers and problem solvers.

Dr. Vijay Shah

Co-founder

Dr. Cong Shen

Co-founder

Mr. Ritesh Gajjar

Co-founder

Dr. Jeffrey H. Reed

Advisor