Blog

GPT-4o and Beyond

by

Today, OpenAI hosted a Spring Update session where they introduced an exciting new model capable of processing audio, vision, and text in real-time. This model is twice as fast as GPT-4-Turbo and costs half as much. We’ve already upgraded the CMA to harness this powerful new model.

Current Use

We’re thrilled to keep enhancing our CRM platform with the latest advancements in Generative AI and Machine Learning. Our customers are already using OpenAI’s Whisper Large-v3 model for multi-lingual translation of call recordings. These transcriptions can be analyzed by GPT-4o and future models to generate concise, high-quality conversational summaries. These summaries help counselors and customer service representatives quickly grasp client interactions, including detailed housing counseling session summaries required for audits and overviews of enrollment and customer service calls. Additionally, OpenAI’s models are being used to evaluate conversations and efficiently handle complex ISO queries, significantly reducing the time required for analysis. A one-hour conversation can now be summarized, analyzed, and reported on within seconds.

AI Task Analysis Engine

Recently, we introduced generative AI capabilities into our Task workflow, integrating AI throughout the business process. This integration allows the analysis of client information across various communication channels like email, text, and phone, helping to securely identify the best-fit programs using financial data. This enables the creation of insights and reports that were previously unattainable with traditional statistical methods. The AI Task Engine also provides real-time recommendations to assist clients effectively.

GPT Agents

We have also introduced GPT Agents outside of the CRM to help build SQL queries for generating custom reports within the CMA. These agents can access the CRM’s database schema and are a step toward natural language reporting. Our GPT Agents can answer questions about system functionality and even process screenshots to provide step-by-step guidance on specific tasks. While these agents are still in the prototype stage and can occasionally produce suboptimal solutions, each advancement in reasoning and memory processing in new models enhances their performance. We are optimistic that these improvements will make interacting with the system easier and faster than ever.

Realtime Audio Processing

We’re also excited about the real-time audio capabilities demonstrated today, which will be available in the coming weeks. We’ve developed Voice Agents to assist with enrollment, qualification, counseling, and customer service. However, these agents were not ready for prime-time as they previously faced delays and pauses as they processed information. The new audio agents are a significant improvement, offering tone and emotional detection, nuanced communication styles, multi-language support, and smooth handling of real-time interruptions. As we offer a comprehensive solution, including phone system hosting, we look forward to integrating these tools to better support your counselors and agents in delivering high-quality assistance to more people.

Beyond

While OpenAI and others work on enhancing these models and their technical infrastructure, we’re committed to exploring how these tools can best support you and your customers. This presents an opportunity to reassess all job roles and discover how these tools can enhance, assist, and redefine our workflow to maximize satisfaction for both our team and those we serve.

We’re excited to unlock the potential of GPT-4o and future models, whether it’s GPT-5 or something entirely new.

Using Spyderbat for our SecOps

by

How Client Support Software uses Spyderbat to secure its cloud native environment

Client Support Software (CSS) is a leading provider of enterprise relationship management solutions for the debt and housing counseling industries. We help our clients streamline their workflows, improve their customer service, and grow their businesses.

As a cloud native company, we rely on Linux VMs and FreeBSD Jails to run our applications and services. Our hosted PBX solution integrates with our CRM software. It runs on Linux and we practice locking access down to reduce the attack surface as much as possible. However, we also face various security challenges in this dynamic and complex environment, such as:

  • Application drift: As we deploy new versions of our software, we need to ensure that they behave as expected and do not introduce any vulnerabilities or performance issues.
  • Supply chain attacks: We use third-party components and libraries in our software development lifecycle (SDLC), which could be compromised by malicious actors or contain hidden backdoors.
  • Zero-day exploits: We need to protect our systems from unknown threats that target unpatched vulnerabilities or exploit novel techniques.

To address these challenges, we partnered with Spyderbat, a cloud native runtime security platform that uses eBPF (extended Berkeley Packet Filter) technology to provide unparalleled visibility and protection for Linux VMs. With over two decades of experience hosting and building Customer Relationship software, we became a Spyderbat Design Partner. This allowed us to provide feedback on use cases that would help tune the product for our usage needs and help those with similar workflows.

Spyderbat is a game changer for us because it gives us the ability to harden our Linux cloud runtime environments and keep our applications rolling. Here are some of the benefits we get from using Spyderbat:

Flashback: Time travel for troubleshooting

Spyderbat’s Flashback feature is like having a continuous runtime digital recorder. It allows us to go back in time and see every step that led to an event of interest, such as a service interruption or operational changes made as a result of a software update.

With Flashback, we can eliminate the pain of scanning logs or reproducing errors. We can instantly pinpoint the root cause of any issue by viewing OS kernel traces that reveal every process, file, network, user, syscall, signal, etc. involved in the causal chain.

Flashback also provides early warning signs of troubling traces by alerting us when it detects anomalies or suspicious behaviors. This helps us proactively prevent problems before they escalate or impact our customers.

Guardian: Application drift detection

Spyderbat’s Guardian feature enables us to reduce interruptions by automatically comparing running applications against prior versions. It alerts us when it detects application drift, which could indicate bugs, performance degradation, or security breaches.

With Guardian, we can end application drift by having the insight to instantly course correct and get our application back on rack.

Interceptor: Signatureless attack prevention

Spyderbat’s Interceptor feature provides automated runtime attack eradication that stops attackers in their tracks using kernel-level eBPF data. It can instantly detect and surgically block problematic traces as they begin to unfold, without relying on signatures or rules.

With Interceptor, we can block attacks targeting known or even unknown vulnerabilities, including:

  • Supply-chain attacks
  • Data exfiltration
  • Malware, Ransomware, and Cryptojacking
  • Zero-Day attacks

Spydertraces: Visual overview of Organization

Sometimes, an image is worth a thousand words. In this case, the causal graphs provided by Spyderbat are worth more than a thousand seconds. Being able to visually see connections and relationships between Linux VMs, processes, and network connections, we are able to quickly get a conceptual grasp of what activity is transpiring throughout our server deployments. This allows us to ask better questions about how we can further secure our infrastructure and saves valuable time by answering existing security questions.

Together, these features allow us to stay ahead of security threats, and focus on providing excellent service to our customers.