Didero is using AI to solve supply chain management at mid-market companies

by with No Comment Tech

Supply chain management concept showing the connections between various order and transport components.

Image Credits: ArtemisDiana / Getty Images

Supply chain management remains a stubborn problem for many mid-market companies that can’t afford SAP or lack sufficient IT resources to manage a complex program. Didero, an early-stage startup, decided to build an AI-fueled tool to make it easier on them.

Today, the company announced a $7 million seed investment. While it was at it, the startup also announced it was emerging from stealth and making its product generally available.

“We are trying to build an end-to-end suite that allows procurement teams to manage their suppliers across a range of existing point solution markets,” Tim Spencer, product lead and co-founder told TechCrunch.

That involves finding suppliers, negotiating contracts, managing purchase orders, producing invoices and making payments, all while providing detailed analysis and handling background supplier management tasks.

Spencer said as they built the product they wanted to take advantage of AI to help their target mid-market companies compensate for their lack of resources. Whereas large companies can force suppliers to play by their rules, smaller companies don’t have that luxury, and AI can handle a lot of the grunt work.

“One of the huge unlocks here, particularly for our wedge of mid-market manufacturers, is AI. Because in the pre-AI world, it was basically not possible to do a lot of these tasks [in an automated way],” Spencer said.

The company uses a variety of AI models, including OpenAI and Google Gemini, depending on the task or requirement, and they are continually experimenting to see which model is best suited to what they are trying to accomplish. “We’re using a lot of foundational models and APIs that are out there. We’re not creating our own foundational models, but we’re doing a lot of fine-tuning of some of the existing models,” company CEO Tom Petit said.

In addition, he said that they have a few very specialized models that they’ve coded themselves to power things like extracting data from tables, purchase orders or price lists — documents that are key to the procurement process.

Spencer and co-founder Lorenz Pallhuber bring the supply chain expertise. Spencer ran procurement at Markai, a startup he helped found, while Pallhuber spent seven years at McKinsey advising Fortune 500 clients on supply chain and procurement software. Petit brings the technical chops and training in AI and machine learning. He also co-founded Landis, a startup that raised over $200 million and helped renters figure out how to get a mortgage.

The company launched in December and they have been building out the product ever since. The $7 million seed round closed last month. The round was led by First Round Capital with participation from Construct Capital, AI Grant, Box Group, Company Ventures and Conviction. Industry angels also contributed.

Sage Geosystems wants to solve the data center energy crisis by storing pressurized water deep underground

by with No Comment Tech

A rig drills into the earth against a cloudy sky.

Image Credits: grandriver / Getty Images

Cindy Taff was standing out in the flat expanse of Starr County, Texas, in early 2022 when she felt it. “It was literally vibrating the ground,” she told TechCrunch. “That was an ‘ah-ha’ moment for me.”

Her startup, Sage Geosystems, was testing equipment used to harvest heat from deep in the earth. The team had injected water into the well and was now letting it back out. The result was a gusher, not of oil, but of hot, clean water that could replace natural gas as a steady source of power throughout the world. 

People have long used the heat deep within the earth, and Sage Geosystems is, too. But it’s also proposing to use wells stretching thousands of feet as batteries, storing water under pressure to generate electricity later. The company has been testing the concept in Starr County for over a year. Sage Geosystems announced Tuesday it’s currently building its first commercial-scale facility just outside of San Antonio.

The new project will occupy most of a 10-acre parcel alongside a coal power plant owned by the San Miguel Electric Cooperative Inc. (known as SMECI). There, Sage plans to drill wells to store electricity from a small solar array and use it to continuously power a small data center, Taff said, calling it  “a model home for a big data center.”

The geopressured geothermal system, as the company calls it, will be rated to produce 3 megawatts of electricity, enough for more than 600 homes, at around 10 cents per kilowatt-hour.

Sage will start drilling in the middle of September, Taff said, and it’ll start the plant in December.

Taff and her colleagues ended up at Sage after long careers in oil and gas. Taff, the company’s CEO, had been at Shell for decades, ultimately as a vice president of onshore drilling. Others had similarly long tenures at Shell, Exxon and elsewhere.

“We wanted to go into renewables,” Taff said, but the transition wasn’t clear cut. Renewables are dominated by wind and solar and don’t overlap much with their skillset, which includes understanding what’s deep in the earth and drilling down to access it. “But when we thought about anything with geology — so energy storage or geothermal — then it was an excellent fit.”

Like many other geothermal startups, Sage Geosystems started with a plan to bring down the cost of electricity. Putting water into the ground is one of the bigger costs that geothermal developers face. Yes, you could wait for water to trickle down the pipe and into the fractured rock around it, but you’d be waiting a while. Instead, they inject the water under pressure, and that takes energy.

When the company started working at its test well, Taff and her colleagues realized they could recoup some of that energy by running the pressurized water through a turbine.

“Basically you’re ballooning the fracture, and you’re storing the water under pressure,” Taff said. “Then when you need it, you basically open a valve on the surface, and that fracture is wanting to close, and it jettisons the water back.”

There, some of the similarities between geothermal and oil and gas start to fade. To frack an oil or gas well, companies inject water and grid (known as a proppant) to crack the rock and keep it open so the fossil fuel can flow back to the well. Much of the water used in drilling is lost, and, oftentimes, briny water emerges alongside the oil and gas. So not only does fracking require lots of water, it also produces plenty of wastewater.

Sage, on the other hand, aims to minimize its water losses. Most happens at the surface when water evaporates from the storage pond. Some more is left when water is pumped from that pond into the well. Taff said that over time, the rock surrounding the well will saturate, forming a barrier that slows losses. When its test well first opened, it lost about 2% to leaks and evaporation for each injection and recovery cycle. A little over a month later, only about 1% was lost per cycle.

Once Sage has proven its technology with the first well, Taff said the company could add up to 10 more wells to bring the site’s capacity up to 50 MW. SMECI, the power cooperative that owns the property, plans to add solar panels at the site in 2026. To provide the kind of consistent power that a coal plant offers, the utility is looking into pairing those panels with some form of energy storage. Overall, the company expects to recover at least 70% of the electricity used to inject the water.

“They want this front-row seat for what we’re doing,” Taff said. “Even though it’s energy storage and not geothermal, it allows us to prove about 80% of our technology.”

Beyond SMECI, Sage is working with big tech companies to develop geothermal and energy storage projects for their data centers. While grid-scale batteries have garnered a lot of attention, they’re a bit too expensive to run a solar-powered data center overnight. 

“We’re not trying to compete with lithium-ion batteries for a two- to three-hour duration because they’ll beat us on cost. But when you have to start stacking lithium ion batteries, we can beat them on cost,” Taff said.

Air Force plane takes off from Larnaca Airport in Cyprus.

Defcon AI closes $44M seed round to solve a problem of 'maximum complexity': Military logistics

by with No Comment Tech

Air Force plane takes off from Larnaca Airport in Cyprus.

Image Credits: Christoph Reichwein/picture alliance / Getty Images

The U.S. Department of Defense is a mammoth organization. It not only employs millions of service members and hundreds of thousands of civilian employees, but also has the world’s largest military budget that’s used to buy and maintain more equipment than can likely fit into a single paragraph. 

It’s a lot to coordinate. Operators within the various agencies of the DOD must make decisions about how to plan their operations, coordinate resources and stay within budget for events that are likely contested — whether that’s from a hurricane or an adversary.

Two years after it was incubated, Virginia-based startup Defcon AI has raised a $44 million seed round led by Bessemer Venture Partners, with participation from Fifth Growth Fund and Red Cell Partners, among others, to solve this seemingly intractable problem.

Consider the Air Mobility Command, a command of the U.S. Air Force. When operators plan airlifts, they have to consider a whole slew of variables: available aircraft, the number of crews required, places for crews to rest, where to refuel, relevant airfields, cargo handling locations. Defcon AI says it has developed a set of software that allows the operator on the front end set these parameters “and then turn the software loose,” Defcon’s co-founder, chief strategy officer and retired U.S. Air Force General Paul Selva told TechCrunch. The software essentially operates against those parameters or inputs to generate the best plan — including the cost tables, resource requirements and schedule. 

This type of planning is difficult enough in the best of circumstances, but during a crisis, defense operators don’t even have the luxury of a day to allocate their resources. That’s where Defcon AI comes in.

“I’ve had all the jobs that we’re actually impacting,” Selva said. During his long military career, Selva held many titles, including the commander of the Air Mobility Command, which oversees nearly all of the Air Force’s fleet of air lift aircraft. He later became the commander of the U.S. Transportation Command, which coordinates transportation missions around the world, including those delivered by ships, trucks, trains and other forms of transportation. Before he retired in 2019, he was nominated by President Barack Obama to be the vice chairman of the Joint Chiefs of Staff. 

He co-founded Defcon in 2022 with Yisroel Brumer and Grant Verstandig, both founding partners of Red Cell Partners (Verstandig is also CEO). Red Cell has an interesting model: The firm makes internal investments but it also incubates companies (including Defcon), often identifying promising entrepreneurs that could lead them. Sometimes, entrepreneurs approach Red Cell before they found a company, and the firm handles things like board building, legal, HR and finance while the company grows. 

In the case of Defcon, Selva says that the company got started “because Air Mobility Command articulated a mission need that wasn’t being filled by industry.” The trio “had a conversation about whether or not we thought this was a tractable problem, and … our intuition was it is a mathematically and software tractable problem, but we have to do it a different way.” 

Brumer and Verstandig have their own impressive pedigrees. Before joining Red Cell, Brumer worked at the Pentagon as acting director of OSD/CAPE (Office of the Secretary of Defense, Cost Assessment and Program Evaluation), an enormous role that essentially functions as the “chief analytics officer” for the DOD, he said, and the overseer for the budget submission process. Verstandig is an entrepreneur who has incubated or grown businesses including Rally Health and defense startup Epirus. 

Defcon AI is targeting a problem of “maximal complexity,” Brumer said. The startup’s system combines different algorithms, including machine learning and mathematical optimization algorithms, to simulate a given scenario and generate the best logistical outcome to meet it. In the initial stages of product development, Defcon used reinforcement learning algorithms that don’t require data, but the company says it is now ingesting more and more data provided by the DOD to power the software. Operators can also choose whether to have the system simulate how an adversary might disrupt the operations, and can tell it to optimize for different variables, like speed versus cost effectiveness.

The company has earned around $15 million in government contracts and delivered a production version that was deployed for a real-world operation with Air Mobility Command less than two years after founding. The company is in the process right now of certifying the software to handle classified, secret information, both to expand its uses in the DOD and to enable it to ingest even more data. It’s also expanding to include trucks, trains and ships to its planning and simulation software.

Defcon is not planning on slowing down. The company sees even more applications across the DOD where its software can make an operational difference, and Brumer said they’re seeing “a very strong demand signal” from the private sector for the product too. Overall, the company says working closely with the end users will result in a better product and a genuine competitive edge in adversarial situations.

“Operational planners are actually trying to assess risk for their commanders,” Selva said. “They’re probably the most skeptical audience for decision support tools, so the extent to which you can partner with them you achieve a better outcome.” 

Planera raises $13.5M to help solve the gnarly problem of scheduling for construction contractors

by with No Comment Tech

Planera co-founders Nitin Bhandari and Saif Lodhi

Image Credits: Planera

Planera, a construction-tech startup that offers scheduling and planning software for commercial construction projects, has raised $13.5 million in a new funding round.

General contractors typically use legacy software such as Oracle Primavera P6 to manage commercial construction projects. These solutions require technical knowledge and do not have an intuitive interface, but contractors still use them for their sophisticated projects that involve multiple milestones and last for some years. For small projects, they sometimes choose a general-purpose tool like Microsoft Project. Nonetheless, these tools are not designed with the requirements of a construction company in mind.

The San Jose-based startup mixes the best of two worlds by offering a standalone software platform that helps contractors schedule and plan their long and short-term projects using a digital whiteboard-like interface, supporting real-time collaborations. Scheduling is crucial for this industry, as construction contracts are quite prone to liquidated damages and have contractual obligations. Contractors need to meet even interim milestones sometimes, and they typically use multiple software to track progress.

With Planera, contractors can collaborate with other contractors involved in the project and get their work done through a single interface.

“The value of Planera is not just in creating the plan but also in being a great communication tool,” said Nitin Bhandari, co-founder and CEO of Planera, in an exclusive interview.

Planera allows general contractors to sketch their construction plans, analyze and optimize them, and schedule them based on their contract terms and obligations. It also includes integrations with commonly used platforms like AutoDesk and Procore so contractors sync their schedules without leaving those apps.

Image Credits: Planera

Bhandari told TechCrunch that the startup plans to expand the software to subcontractors as general contractors often work with subcontractors for specialty work, such as electrical and mechanical.

“As construction companies are modernizing other aspects of their business and getting the benefits of that, now they’re like, hey, we need to modernize all critical aspects of the business, and scheduling and planning definitely is now part of the focus,” he stated.

There are plenty of construction-related startups, but Planera’s founding team has had some success in the past. Bhandari previously co-founded mobile browser company Skyfire and screen-time app ZenScreen, which was later acquired by Life360, before starting with Planera in November 2021.

Bhandari got the idea to start Planera after meeting Saif Lodhi, who has been running a general contractor company, California Engineering Contractors, for about 30 years.

Initially, Lodhi asked Bhandari to help him modernize the contractor company. However, after spending a couple of months, Bhandari discovered that “scheduling was completely broken” in the construction industry. He brought in Erik Swenson who was the CTO at Skyfire for more than 10 years, to help him establish Planera to solve the problem.

Cut to today, Planera serves more than two dozen customers that use its software on over 500 live projects.

Planera’s all-equity Series A round was led by Sierra Ventures, along with participation from Sorenson Capital, Brick and Mortar Ventures, Prudence VC and Firebolt Ventures.

Bhandari told TechCrunch that 60% of its fresh funding would be used to invest more in sales and marketing to amplify the startup’s go-to-market. The remaining 40% would be deployed in R&D and product development. The startup also plans to integrate construction-specific AI to bring efficiency to scheduling and planning.

“We will be building our own data models and assistant-like functionality, which will start rolling out toward the end of the year and early part of next year,” he said.

Bhandari did not share Planera’s overall revenues, though he said revenues will be 5-8X higher compared to its Q4 of last year. The remote-culture startup also has a 30-person workforce across markets, including some in India’s Bengaluru, and plans to expand to 45 or 50 in the next six months.

Sage Geosystems wants to solve the data center energy crisis by storing pressurized water deep underground

by with No Comment Tech

A rig drills into the earth against a cloudy sky.

Image Credits: grandriver / Getty Images

Cindy Taff was standing out in the flat expanse of Starr County, Texas, in early 2022 when she felt it. “It was literally vibrating the ground,” she told TechCrunch. “That was an ‘ah-ha’ moment for me.”

Her startup, Sage Geosystems, was testing equipment used to harvest heat from deep in the earth. The team had injected water into the well and was now letting it back out. The result was a gusher, not of oil, but of hot, clean water that could replace natural gas as a steady source of power throughout the world. 

People have long used the heat deep within the earth, and Sage Geosystems is, too. But it’s also proposing to use wells stretching thousands of feet as batteries, storing water under pressure to generate electricity later. The company has been testing the concept in Starr County for over a year. Sage Geosystems announced Tuesday it’s currently building its first commercial-scale facility just outside of San Antonio.

The new project will occupy most of a 10-acre parcel alongside a coal power plant owned by the San Miguel Electric Cooperative Inc. (known as SMECI). There, Sage plans to drill wells to store electricity from a small solar array and use it to continuously power a small data center, Taff said, calling it  “a model home for a big data center.”

The geopressured geothermal system, as the company calls it, will be rated to produce 3 megawatts of electricity, enough for more than 600 homes, at around 10 cents per kilowatt-hour.

Sage will start drilling in the middle of September, Taff said, and it’ll start the plant up in December.

Taff and her colleagues ended up at Sage after long careers in oil and gas. Taff, the company’s CEO, had been at Shell for decades, ultimately as a vice president of onshore drilling. Others had similarly long tenures at Shell, Exxon and elsewhere.

“We wanted to go into renewables,” Taff said, but the transition wasn’t clear cut. Renewables are dominated by wind and solar and don’t overlap much with their skillset, which includes understanding what’s deep in the earth and drilling down to access it. “But when we thought about anything with geology — so energy storage or geothermal — then it was an excellent fit.”

Like many other geothermal startups, Sage Geosystems started with a plan to bring down the cost of electricity. Putting water into the ground is one of the bigger costs that geothermal developers face. Yes, you could wait for water to trickle down the pipe and into the fractured rock around it, but you’d be waiting a while. Instead, they inject the water under pressure, and that takes energy.

When the company started working at its test well, Taff and her colleagues realized that they could recoup some of that energy by running the pressurized water through a turbine.

“Basically you’re ballooning the fracture, and you’re storing the water under pressure,” Taff said. “Then when you need it, you basically open a valve on the surface, and that fracture is wanting to close, and it jettisons the water back.”

There, some of the similarities between geothermal and oil and gas start to fade. To frack an oil or gas well, companies inject water and grid (known as a proppant) to crack the rock and keep it open so the fossil fuel can flow back to the well. Much of the water used in drilling is lost, and oftentimes, briny water emerges alongside the oil and gas. So not only does fracking require lots of water, it also produces plenty of wastewater.

Sage, on the other hand, aims to minimize its water losses. Most happens at the surface when water evaporates from the storage pond. Some more is left when water is pumped from that pond into the well. Taff said that over time, the rock surrounding the well will saturate, forming a barrier that slows losses. When its test well first opened, it lost about 2% to leaks and evaporation for each injection and recovery cycle. A little over a month later, only about 1% was lost per cycle.

Once Sage has proven its technology with the first well, Taff said the company could add up to 10 more wells to bring the site’s capacity up to 50 MW. SMECI, the power cooperative that owns the property, plans to shutter its coal plant at the site in 2026, replacing it with solar panels. To provide the kind of consistent power that a coal plant offers, the utility is looking into pairing those panels with some form of energy storage. Overall, the company expects to recover at least 70% of the electricity used to inject the water.

“They want this front row seat for what we’re doing,” Taff said. “Even though it’s energy storage and not geothermal, it allows us to prove about 80% of our technology.”

Beyond SMECI, Sage is working with big tech companies to develop geothermal and energy storage projects for their data centers. While grid-scale batteries have garnered a lot of attention, they’re a bit too expensive to run a solar-powered data center overnight. 

“We’re not trying to compete with lithium-ion batteries for a two- to three-hour duration because they’ll beat us on cost. But when you have to start stacking lithium ion batteries, we can beat them on cost,” Taff said.

Didero is using AI to solve supply chain management at mid-market companies

by with No Comment Tech

Supply chain management concept showing the connections between various order and transport components.

Image Credits: ArtemisDiana / Getty Images

Supply chain management remains a stubborn problem for many mid-market companies that can’t afford SAP or lack sufficient IT resources to manage a complex program. Didero, an early-stage startup, decided to build an AI-fueled tool to make it easier on them.

Today, the company announced a $7 million seed investment. While it was at it, the startup also announced it was emerging from stealth and making its product generally available.

“We are trying to build an end-to-end suite that allows procurement teams to manage their suppliers across a range of existing point solution markets,” Tim Spencer, product lead and co-founder told TechCrunch.

That involves finding suppliers, negotiating contracts, managing purchase orders, producing invoices and making payments, all while providing detailed analysis and handling background supplier management tasks.

Spencer said as they built the product they wanted to take advantage of AI to help their target mid-market companies compensate for their lack of resources. Whereas large companies can force suppliers to play by their rules, smaller companies don’t have that luxury, and AI can handle a lot of the grunt work.

“One of the huge unlocks here, particularly for our wedge of mid-market manufacturers, is AI. Because in the pre-AI world, it was basically not possible to do a lot of these tasks [in an automated way],” Spencer said.

The company uses a variety of AI models, including OpenAI and Google Gemini, depending on the task or requirement, and they are continually experimenting to see which model is best suited to what they are trying to accomplish. “We’re using a lot of foundational models and APIs that are out there. We’re not creating our own foundational models, but we’re doing a lot of fine-tuning of some of the existing models,” company CEO Tom Petit said.

In addition, he said that they have a few very specialized models that they’ve coded themselves to power things like extracting data from tables, purchase orders or price lists — documents that are key to the procurement process.

Spencer and co-founder Lorenz Pallhuber bring the supply chain expertise. Spencer ran procurement at Markai, a startup he helped found, while Pallhuber spent seven years at McKinsey advising Fortune 500 clients on supply chain and procurement software. Petit brings the technical chops and training in AI and machine learning. He also co-founded Landis, a startup that raised over $200 million and helped renters figure out how to get a mortgage.

The company launched in December and they have been building out the product ever since. The $7 million seed round closed last month. The round was led by First Round Capital with participation from Construct Capital, AI Grant, Box Group, Company Ventures and Conviction. Industry angels also contributed.

Diadem Capital, Joe Hammill, Stephanie Rieben, venture capital

Two ex-Wall Streeters want to solve one of VC’s biggest problems: Warm introductions

by with No Comment Tech

Diadem Capital, Joe Hammill, Stephanie Rieben, venture capital

Image Credits: Diadem Capital / Diadem Capital co-founders Joe Hammill and Stephanie Rieben

Diadem Capital is throwing its hat into the crowded space of making funding more accessible and easy to obtain for high-growth startups. And it promises your next round will close “5x faster.”

Buoyed by a $600,000 pre-seed round led by Launch NY, the Buffalo-based fundraising platform, which touts itself as a “warm introduction network,” is building a company, investor and lending matching program in a similar vein to platforms like SeedInvest.

Diadem’s co-founders Stephanie Rieben and Joe Hammill started the company two years ago after a decade in investment banking, capital markets and trading on Wall Street. They parted ways for a bit before reuniting at Hum Capital, a venture debt funding platform that matched companies and lenders.

While at Hum, Rieben and Hammill spoke to founders who needed equity, but were too early or not in the position to give up a percentage of their business.

“That’s when we started talking about doing something about that,” CEO Rieben told TechCrunch.

3 years after BLM, here’s who stuck to their diversity commitments

Here’s what they created: The company built a low-code platform where founders sign up for capital. Those applications are vetted by Rieben and Hammill, who then personally meet with founders they want to work with. Once founders are accepted onto the platform, they are matched with institutional investors.

To be eligible, companies must be VC-backed and have at least $1 million in annual recurring revenue. On the debt side, the company will help at all levels, including bootstrapped companies, that have up to $50 million in ARR. The company has future plans to increase that to $100 million ARR, Rieben said.

Founders can view their progress in terms of who they were introduced to and the status of those relationships. Investors can come directly onto the platform, but are not able to see their deal flow until they meet with Rieben as a way to reduce friction and make sure investors will follow through on introductions.

“We check in with both the founder and investor,” Hammill said. “Founders never get real feedback because the system is not set up to do that. As an intermediary, we’re building out a place where the investor can give true honest feedback that is aggregated. For example, when three investors submit a review on the first call, that’s when the founder gets that data and individual feedback in an aggregated way.”

Four venture capital personas (and how to land them)

At a time when progress has been slow to fund underrepresented founders, Hammill defended Diadem’s VC-backed strategy, saying that while the company does want to help all founders, it isn’t helping with pitch decks or the pitch itself.

“We will not turn down someone who’s not VC-backed and who’s running an amazing business,” he said. “That said, pitching is part science and part art. We don’t shy away from that, but we prefer them to be VC-backed because that at least shows to us that they know how to pitch and close a VC.”

Meanwhile, Diadem currently has more than 100 lenders on its platform and over 800 venture capitalists using its platform. So far, over 1,500 startups have applied, and 17 founders have raised more than $60 million total. Typically, a fundraise takes four to six months on the equity side, however, Diadem has been able to reduce that time down to two to three months, Rieben said.

The pair was mum on how much revenue Diadem has brought in other than saying the company is posting revenue currently. And unlike other competitors, Rieben and Hammill are licensed bankers, so they charge a success-based fee model.

“We’re very different from competitors that have like SaaS models or a pay-to-play model where early-stage founders have to pay like $5,000 a month for six months,” Rieben said. “Many times, they get no investor introductions, or very little, or don’t get funded. We’re focused on fundability.”

8 reasons why the venture capital market isn’t as miserable as you think

Diadem Capital, Joe Hammill, Stephanie Rieben, venture capital

Two ex-Wall Streeters want to solve one of VC’s biggest problems: Warm introductions

by with No Comment Tech

Diadem Capital, Joe Hammill, Stephanie Rieben, venture capital

Image Credits: Diadem Capital / Diadem Capital co-founders Joe Hammill and Stephanie Rieben

Diadem Capital is throwing its hat into the crowded space of making funding more accessible and easy to obtain for high-growth startups. And it promises your next round will close “5x faster.”

Buoyed by a $600,000 pre-seed round led by Launch NY, the Buffalo-based fundraising platform, which touts itself as a “warm introduction network,” is building a company, investor and lending matching program in a similar vein to platforms like SeedInvest.

Diadem’s co-founders Stephanie Rieben and Joe Hammill started the company two years ago after a decade in investment banking, capital markets and trading on Wall Street. They parted ways for a bit before reuniting at Hum Capital, a venture debt funding platform that matched companies and lenders.

While at Hum, Rieben and Hammill spoke to founders who needed equity, but were too early or not in the position to give up a percentage of their business.

“That’s when we started talking about doing something about that,” CEO Rieben told TechCrunch.

3 years after BLM, here’s who stuck to their diversity commitments

Here’s what they created: The company built a low-code platform where founders sign up for capital. Those applications are vetted by Rieben and Hammill, who then personally meet with founders they want to work with. Once founders are accepted onto the platform, they are matched with institutional investors.

To be eligible, companies must be VC-backed and have at least $1 million in annual recurring revenue. On the debt side, the company will help at all levels, including bootstrapped companies, that have up to $50 million in ARR. The company has future plans to increase that to $100 million ARR, Rieben said.

Founders can view their progress in terms of who they were introduced to and the status of those relationships. Investors can come directly onto the platform, but are not able to see their deal flow until they meet with Rieben as a way to reduce friction and make sure investors will follow through on introductions.

“We check in with both the founder and investor,” Hammill said. “Founders never get real feedback because the system is not set up to do that. As an intermediary, we’re building out a place where the investor can give true honest feedback that is aggregated. For example, when three investors submit a review on the first call, that’s when the founder gets that data and individual feedback in an aggregated way.”

Four venture capital personas (and how to land them)

At a time when progress has been slow to fund underrepresented founders, Hammill defended Diadem’s VC-backed strategy, saying that while the company does want to help all founders, it isn’t helping with pitch decks or the pitch itself.

“We will not turn down someone who’s not VC-backed and who’s running an amazing business,” he said. “That said, pitching is part science and part art. We don’t shy away from that, but we prefer them to be VC-backed because that at least shows to us that they know how to pitch and close a VC.”

Meanwhile, Diadem currently has more than 100 lenders on its platform and over 800 venture capitalists using its platform. So far, over 1,500 startups have applied, and 17 founders have raised more than $60 million total. Typically, a fundraise takes four to six months on the equity side, however, Diadem has been able to reduce that time down to two to three months, Rieben said.

The pair was mum on how much revenue Diadem has brought in other than saying the company is posting revenue currently. And unlike other competitors, Rieben and Hammill are licensed bankers, so they charge a success-based fee model.

“We’re very different from competitors that have like SaaS models or a pay-to-play model where early-stage founders have to pay like $5,000 a month for six months,” Rieben said. “Many times, they get no investor introductions, or very little, or don’t get funded. We’re focused on fundability.”

8 reasons why the venture capital market isn’t as miserable as you think

documents, title, startup, venture capital

Why RAG won't solve generative AI's hallucination problem

by with No Comment Tech

documents, title, startup, venture capital

Image Credits: D3Damon / Getty Images

Hallucinations — the lies generative AI models tell, basically — are a big problem for businesses looking to integrate the technology into their operations.

Because models have no real intelligence and are simply predicting words, images, speech, music and other data according to a private schema, they sometimes get it wrong. Very wrong. In a recent piece in The Wall Street Journal, a source recounts an instance where Microsoft’s generative AI invented meeting attendees and implied that conference calls were about subjects that weren’t actually discussed on the call.

As I wrote a while ago, hallucinations may be an unsolvable problem with today’s transformer-based model architectures. But a number of generative AI vendors suggest that they can be done away with, more or less, through a technical approach called retrieval augmented generation, or RAG.

Here’s how one vendor, Squirro, pitches it:

At the core of the offering is the concept of Retrieval Augmented LLMs or Retrieval Augmented Generation (RAG) embedded in the solution … [our generative AI] is unique in its promise of zero hallucinations. Every piece of information it generates is traceable to a source, ensuring credibility.

Here’s a similar pitch from SiftHub:

Using RAG technology and fine-tuned large language models with industry-specific knowledge training, SiftHub allows companies to generate personalized responses with zero hallucinations. This guarantees increased transparency and reduced risk and inspires absolute trust to use AI for all their needs.

RAG was pioneered by data scientist Patrick Lewis, researcher at Meta and University College London, and lead author of the 2020 paper that coined the term. Applied to a model, RAG retrieves documents possibly relevant to a question — for example, a Wikipedia page about the Super Bowl — using what’s essentially a keyword search and then asks the model to generate answers given this additional context.

“When you’re interacting with a generative AI model like ChatGPT or Llama and you ask a question, the default is for the model to answer from its ‘parametric memory’ — i.e., from the knowledge that’s stored in its parameters as a result of training on massive data from the web,” David Wadden, a research scientist at AI2, the AI-focused research division of the nonprofit Allen Institute, explained. “But, just like you’re likely to give more accurate answers if you have a reference [like a book or a file] in front of you, the same is true in some cases for models.”

RAG is undeniably useful — it allows one to attribute things a model generates to retrieved documents to verify their factuality (and, as an added benefit, avoid potentially copyright-infringing regurgitation). RAG also lets enterprises that don’t want their documents used to train a model — say, companies in highly regulated industries like healthcare and law — to allow models to draw on those documents in a more secure and temporary way.

But RAG certainly can’t stop a model from hallucinating. And it has limitations that many vendors gloss over.

Wadden says that RAG is most effective in “knowledge-intensive” scenarios where a user wants to use a model to address an “information need” — for example, to find out who won the Super Bowl last year. In these scenarios, the document that answers the question is likely to contain many of the same keywords as the question (e.g., “Super Bowl,” “last year”), making it relatively easy to find via keyword search.

Things get trickier with “reasoning-intensive” tasks such as coding and math, where it’s harder to specify in a keyword-based search query the concepts needed to answer a request — much less identify which documents might be relevant.

Even with basic questions, models can get “distracted” by irrelevant content in documents, particularly in long documents where the answer isn’t obvious. Or they can — for reasons as yet unknown — simply ignore the contents of retrieved documents, opting instead to rely on their parametric memory.

RAG is also expensive in terms of the hardware needed to apply it at scale.

That’s because retrieved documents, whether from the web, an internal database or somewhere else, have to be stored in memory — at least temporarily — so that the model can refer back to them. Another expenditure is compute for the increased context a model has to process before generating its response. For a technology already notorious for the amount of compute and electricity it requires even for basic operations, this amounts to a serious consideration.

That’s not to suggest RAG can’t be improved. Wadden noted many ongoing efforts to train models to make better use of RAG-retrieved documents.

Some of these efforts involve models that can “decide” when to make use of the documents, or models that can choose not to perform retrieval in the first place if they deem it unnecessary. Others focus on ways to more efficiently index massive datasets of documents, and on improving search through better representations of documents — representations that go beyond keywords.

“We’re pretty good at retrieving documents based on keywords, but not so good at retrieving documents based on more abstract concepts, like a proof technique needed to solve a math problem,” Wadden said. “Research is needed to build document representations and search techniques that can identify relevant documents for more abstract generation tasks. I think this is mostly an open question at this point.”

So RAG can help reduce a model’s hallucinations — but it’s not the answer to all of AI’s hallucinatory problems. Beware of any vendor that tries to claim otherwise.