CalAster

Pandemic outbreaks and natural disasters cause sudden upticks in emergency calls and saturation of Public Safety Answering Points (PSAP) that are currently impossible to handle quickly. This has once again proven to be true in the recent COVID19 worldwide pandemic. CalAster’s core technology is designed to help emergency call takers efficiently handle situations of saturation through the classification and featurization of incoming 911 emergency calls. Additionally, CalAster is developing a simulation tool to measure the impact of prioritization policies on emergency response times in periods of saturation. Such a tool will be critical for decision-makers to strategize resource distributions and correctly prepare for current and future emergency crises.

Every year, more than 240 million emergency calls are made in the US. Operators have to manually collect information about each emergency while discarding junk information. In fact, it has been reported that roughly 15 to 20% of inbound 911 calls are non-emergencies [1]. This painstaking process wastes precious time and resources. In addition, Public Safety Answering Points (PSAPs) are ill-equipped to respond to sudden upticks in calls during mass casualty events, as has been shown once again by the recent COVID-19 pandemic [2]. Whereas PSAP saturation has been more serious than ever all over the country, large cities such as New York City have been particularly hit. This saturation has massive societal consequences at a large scale: a couple of lost seconds can be life-changing for many victims. Coupled with a high-stress environment which causes frequent post traumatic stress disorders (PTSD) and understaffing, today’s emergency call centers have room for improvement [3][4]. 

[1] Percentage of non-emergency calls: https://www.ncjrs.gov/pdffiles...

[2] NY Times Headlines regarding 911 Saturation linked to COVID-19: https://www.nytimes.com/2020/0...

[3] 911 Report from Vera Institute of Justice: https://www.vera.org/publicati...

[4] Quantifying the impact of emergency response times: https://info.rapidsos.com/down...

CalAster is an innovative solution designed to analyze, prioritize, and visualize emergency calls using state-of-the-art machine learning algorithms. By plugging ourselves onto PSAPs, we handle inbound audio streams, effectively use real-time speech-to-text to retrieve the transcripts and exploit natural language processing to extract key features from incoming emergency calls. The sensitivity of the information we process is one of our top priorities and we run on a scalable HIPAA/GDPR-compliant cloud infrastructure to match the need for redundancy and security. This innovative solution will be the first layer of a new generation of machine intelligent emergency response systems. CalAster also helps call takers make faster decisions given a limited set of variables through 911 call featurization and prioritization. 

Aside from supporting call takers, CalAster also helps decision-makers make fast decisions in crisis situations by providing an access to real-time and large-scale monitoring of emergency calls. CalAster is currently developing a simulation model to quantify the impact of smart interactive voice response (IVR) and automated filtering, and identify further limitations of the current infrastructures while proposing data-driven solutions. These two goals allow CalAster to broaden its impact, working simultaneously on the “how” and the “why” of emergency calls filtering and prioritization.

CalAster’s solution directly impacts the daily activities of 911 call operators. Starting with the I-Corps program, we interviewed dozens of dispatchers, call takers, PSAP supervisors, policy makers, and others. Our goals were two-fold. First, to precisely understand the challenges faced every day by those who dedicate their lives to saving the one of others. Second, to discuss and evaluate the quality of potential solutions, their feasibility and the impact they would have on the field.

CalAster’s indirect impact, however, is all over the emergency response chain. We help call takers handle stressful situations and decision makers with big data, which ultimately helps ground units target their limited resources, and results in less victims dying from their critical conditions. 

Last but not least, CalAster plans to give back to the overall community by open-sourcing its technology. This will allow disadvantaged call centers around the world to easily recreate call analysis pipelines and continue CalAster’s mission in all areas of the globe. 

CalAster enhances medical emergency victims support and improves pandemic data collection. On the one hand, its incoming call classification and featurization tool will allow for extremely fast and global symptom hotline deployments. Victims will start by describing their symptoms in their preferred language over the phone. CalAster will then automatically link them to the best medical match based on their description and language: doctor, specialist, online resource, or even volunteering medical student for less critical emergencies. On the other hand, decision makers will have access to unprecedented, country-wide data that will allow better management of global pandemics.

Policy makers have been increasingly conscious of how outdated the current emergency infrastructure is, be it in the USA or elsewhere in the world. Even more so now that the world was hit by the COVID19 pandemic, we see governments moving towards integrating new technologies to replace the legacy 911 architectures, while ensuring a broader service interoperability. Three players have been making a fuss in that space recently: RapidSOS, RapidDeploy and Carbyne. They exploit technology on the cloud to patch the legacy 911 architecture and offer new perspectives to first responders: better UX and UI, streamlined communication without the radio, centralized data architectures, and localization through our smartphones GPS. Nonetheless, none of these emergent players play on NG911, the next generation of 911 technology, missing the value of IP-based communication, and are quoted as ‘temporary solutions’ by the stakeholders we have had the chance to interview. Because we are coming on top of any architecture, we support first responders on legacy architecture, and pave the way by showing the potential of VoIP based emergency response.

At the macro-scale, our core technology is two-fold: machine-learning for inbound emergency calls, and cloud architecture for virtual queueing and emergency management. 

In the first case, we base our technology on VoIP to manage audio streams and use our containerized AI models through multiple microservices, such as speech-to-text, background noise classification, gender estimation, age prediction, stress quantification, and emotion and tone extraction. That first step provides feature engineering, while a second layer of machine learning models determine the type of emergency, and consolidate the featurization by merging the pieces together.

On the other hand, we have the cloud architecture we designed as a smart interactive voice response system, that can be spawned when saturation hits a call center. Its purpose is to enable call centers to take all incoming calls and extract a minimum amount of information while redispatching the call operators according to the information collected. It finally keeps each call ‘alive’ unless the caller hangs up, in order to accelerate the process and enable the call to be diverted to the first team available on-the-fly. The latter works as a standalone, and is autonomous throughout the cloud architecture we built, while the first solution is used as a modular endpoint, integrated in existing computer aided-dispatch softwares, already in use in PSAPs.

Similar technologies are already exploited in customer call centers, with the purpose of quantifying KPIs on the fly for their users. Some are deprecated (DTMF-based IVRs, e.g. IRS), while others are slowly overcrowding the market of marketing-centered call centers (would it be for CRM purposes or customer success more specifically). Regarding the endpoint modularity architecture, Corti, a danish competitor, focuses on stroke detection over the phone in Europe. They successfully plugged in their API on existing CADs in Denmark. Enhanced information streams and better UX is also now getting widely accepted through companies such as RapidSOS, RapidDeploy and Carbyne.

While we sought to understand the field better, we stumbled upon the fact a majority of the problems in the field could have been easily avoided if properly thought twice. The scalability of the architecture during sudden upticks in calls (mass casualty events such as the COVID19 pandemics being one good illustration), the interoperability of the product call takers use, the lacking system engineering to determine the impact on social inequalities and concrete health metrics,... All of those could benefit from a rewiring. There is also little to no feedback loop between the emergency response actors and the policy makers… Knowing how the industry works right now has led us to define sustainable long-term goals as part of the technology we are building.

Following the Lean Startup strategy, the 50+ interviews we conducted led us to our theory of change, or as we frame it, the need for change: making inbound calls interoperable through our virtual queue infrastructure, both avoiding losing information and enabling more people to participate in the effort to process all of them through an elastic resource management; bridging the infrastructure between counties (e.g. NENA new standards); providing real-time data pattern analysis through centralized databases, fed into factor analysis framework for emergency response times; reducing training time by bringing what technology has been best for the past decades, aka good UI and good UX (from years to hours, 10-8 Systems being one of the examples), which also has an impact on turnover and PTSD; helping call operators to be understanding the calls faster and better, while dispatching the right resources at the right time [1] (comes along with the most recent BLM movement).

[1] San Francisco police not responding to non-criminal emergencies anymore

CalAster has a POC ready and is currently developing an MVP while developing partnerships to start pilots. The latter will allow CalAster’s technology to spread out and demonstrate the credibility and safety of the technology’s added value to major actors in the field. As CalAster’s solution interacts both with victims and professional operators, we ultimately impact different populations.

In a year, we aim to implement at least two pilots: one in the US and one in France. With these pilots, CalAster will serve around 100 PSAP operators and will cover a minimum of 100k inhabitants, based on the PSAP’s location. Meanwhile, we do give lots of importance to partnerships with organizations such as NENA and EENA, paving the way for the next generation of emergency response while also creating whole new standards. This is to us the only way to work hand-in-hand with all the actors involved, would they be private, public, governmental or consultants.

In five years, we aim to be running in around 50 large PSAPs, serving around 1500 operators and covering the population in the respective locations, accounting to around 10 million people. Our models will have undergone intense testing and certification by public organizations in the US and Europe to provide complete support to dispatchers. We appreciate that the market in the emergency response field can be slow at times in regards to changes and partnership can take a long time to finalize.

Within the next year, our main goal is to develop pilots in at least two locations. We are currently discussing with potential partners in the US and in France interested in developing machine learning solutions for their CAD software. Pilots represent the key to reach fresh data from call centers and the best way to improve technical AI models.

To multiply the impact of CalAster in the emergency sector, we will require certifications. We are aiming to develop a transparent, secure and accessible solution to obtain the accreditation of organizations such as NENA in the US and EENA in Europe. Dealing with these two is the prime goal to pursue and entrench CalAster in the emergency response field.

Over the next five-years, our goal is to implement CalAster’s technology in 20 to 50 call centers in the US and Europe while covering a wide diversity of PSAPs. Diversity is key: replicating a data-based solution from one location to another is nearly impossible given the large heterogeneity in PSAPs across the country. Achieving diversity will allow successful scalability in future deployments. We plan to become an accomplished actor in the next generation of technology providers in the emergency response field.

The biggest challenge CalAster faces in the next year is the closure of the right data partnership with an established call center. The data collected from the calls will allow CalAster to iterate on its product and test its model in real-world conditions. This initial partnership will serve as pilote test for CalAster.

In the next five years, the biggest barrier to CalAster’s success will be scaling and expanding to other call centers. The fragmented emergency response ecosystem will make it difficult at first to establish CalAster as the gold standard for all call centers. There is a small yet undeniable cost to starting to use CalAster - staff needs to be trained and budget needs to be allocated. This will require convincing local governments which may be slow and painstaking. To alleviate this concern, CalAster is actively looking for partnerships which could include multiple call centers at once. This will provide CalAster with sufficient data to later convince individual call centers. In that regard, leveraging CalAster’s connection to the NENA or the American Red Cross could be a way to achieve scale with a lower sales effort. 

Data Partnership: this key barrier is paramount in the tight space of emergency response and is based on trust between the call center and CalAster. The first step to create trust will be to work closely with established CAD providers and, after demonstrating the technical efficacy of our technology, count on them to give CalAster an additional layer of credibility. It will be easier to demonstrate the robustness of CalAster’s solution to technical professionals, rather than to PSAP directors. We have already initiated such a partnership in France and are now willing to do the same in the US.

Scaling and expanding: we believe that a strong network and support system will be crucial. We will have to recruit and work with well-connected 911 professionals to lower the barrier to entry in new PSAPs. Our team is expanding its network in the emergency response field, but we believe that having employees with on-the-field 911 experience present the product to CADs will be extremely valuable to achieve additional credibility and relate CalAster’s solution to real-life experiences. We believe we will gain significant momentum after successfully completing the first pilots. At this point, the fragmented PSAP ecosystem could become an asset due to the accessibility of our process and the time spent on diversification during pilot tests.

4 co-founders, working part time on the solution.

Our team combines experience in the healthcare sector with an expertise in machine learning. This stands the team in good stead to understand the problems faced by emergency call centers and build a viable long term solution.

In addition, the team is already in contact with some of the most influential organizations in the emergency sector including NENA (National Emergency Number Association) and the Red Cross in the US, and Atraksis and EENA (European Emergency Number Association) in Europe. This puts Calaster in a privileged position to bring forward meaningful change through partnerships with all of these organizations.

Overall emergency response time, time to reach an operator on the phone, and staff turnaround are common KPIs across call centers. CalAster can help public safety answering points achieve all of these goals. CalAster’s priority is to reduce the amount of time operators need to spend on each call. This reduces call center saturation and therefore ensures victims can get on the line with an operator faster. By reducing the workload for call takers, CalAster reduces the amount of stress at work which is one of the main factors behind the high staff turnaround. Ultimately, by reducing staff turnaround and the need for hiring additional call takers, Calaster can help bring down the maintenance cost of public safety answering points.

CalAster’s streamlining of emergency call analysis provides value for victims who get help faster, and emergency call centers which have lower operational costs. This dual argument will appeal to the decision makers, often local politicians, who are focused on improving the quality of life of their constituents whilst maintaining a tight budget.

CalAster is currently funded by its founders. The software only approach and the fact that the founders do not receive any salary means little capital is needed to deploy an MVP. The team will still apply for grants in the upcoming months in order to recruit new team members to develop the tech stack faster, and cover legal fees associated with the data partnerships which the company will sign with call centers as part of the initial development phase.

Once the proof of concept has been established and CalAster has been used in call centers, the company will look into raising investment capital to further fund growth, both in the US and in Europe. Hiring of sales people will be critical given the fragmented nature of the emergency response ecosystem.

Long term, CalAster’s revenue will be derived from emergency call centers subscriptions. Counties using CalAster will pay a monthly fee which will be a function of the call center size and the number of calls coming in daily. In addition, an initial setup transaction fee will be charged to install the system in the call centers and provide training to the call operators.

CalAster is developing well and, thanks to a growing network in the emergency response field, has access to some valuable mentorship. But applying to solve is an amazing opportunity for CalAster for three reasons:

1. Mentorship: being assigned a formal mentor that is knowledge and well-connected in the 911 space would allow us to save significant time in network development and speed up the implementation of CalAster’s first pilots, which will be crucial for the development of the rest of the project. As was mentioned earlier in the application, the 911 space is tight and credibility matters immensely. Without a name, one has low chances of long-term survival in a space that is desperately needing a fresh wave of new technologies.

2. Publicity: Solve’s platform is an excellent way of reaching out to a network larger than we possibly could have done ourselves, while making the public more familiar with our technology. What matters to us is helping the community, and the more people start to understand how PSAPs can be improved the better. PSAP call operators are at the forefront of any emergency and they often do not get all the credit they deserve.

3. Funding: while we have been able to bootstrap the venture thus far, expenses quickly pile up. Being financially supported would allow us to cover legal fees, hire new members, continue working on the tech stack and travel to public events to pitch our solution and increase our reach.

At this stage, CalAster is looking for partnerships in the US with three types of partners:

1. Mentors or advisors familiar with the emergency response industry. Their guidance, help, and network will be invaluable.

2. Public Safety Answering Points: key partnerships to get access to real-world data and improve the technology

3. Computer-Aided Dispatch Software Organizations: key partnerships to provide a platform that will allow for a seamless integration of our solution into PSAPs for pilot tests.