The Evidence Has to Come From Somewhere

Legora

Legora's first paying customer was a friend's mother. She paid $100 a month, according to CEO and co-founder Max Junestrand's interview with Pigment. A person with legal work to finish gave the business its first recurring reason to exist.

That relationship now has institutional company. Legora announced a $550M Series E on October 21 at an $8.8B post-money valuation, led by Atreides Management. Baron Capital, Capital Group, Fidelity Management & Research Company, Glade Brook Capital Partners and J.P. Morgan joined the round.

Returning investors include Accel, Benchmark, Bessemer Venture Partners, General Catalyst, ICONIQ, Insight Venture Partners, Menlo Ventures, Redpoint Ventures, Salesforce Ventures and Sands Capital. The investor list has grown into quite a meeting. The assignment still starts with a lawyer trying to get useful work done.

Max Junestrand built Legora with Sigge Labor and August Erséus. Sigge Labor is now president; Jacob Lauritzen is CTO. In a Bain Capital Ventures interview, Max Junestrand described moving into a Swedish law firm's offices and eating lunch with its lawyers. Customer research came with colleagues at the table.

Legora now combines research, document review, drafting and multi-step agents in its legal operating system. The company reports more than $200M ARR and use across more than 2,400 law firms and in-house teams. Those figures describe management's reported scale, with the software reaching both the people billing for legal work and the people managing the legal budget.

That mix matters. In-house teams represented more than 40% of new customers in Q3, Legora says. A firm can use additional capacity to serve clients; a corporate legal department has to decide where that capacity belongs inside its own business. The same vendor has to understand both conversations. For legal operations leaders, a rollout also means changing who owns the next step.

The new capital supports expansion across North America, Europe and Asia-Pacific, alongside more investment in Legora aOS. Its CFO, David Eckstein, has put durable revenue and a stronger cost structure inside the company's stated ambition. Supporting adoption across more offices and more kinds of legal teams becomes part of delivering on it.

For the team that once learned over lunch inside a customer's firm, the growing job is to keep that proximity as the next legal department brings its own documents, expectations and way of working through the door.

MARC

Somewhere in a property manager's files is a document giving the owner permission to challenge the next bill. MARC wants that agreement available when somebody can still do something about the charge.

Aaron Devitt, founder and CEO, has raised a $5M Seed round led by Work-Bench, with MXV Capital participating. Announced October 22, the financing brings reported total funding to $6M and backs expansion across the US and Canada, plus engineering and commercial hiring in New York.

Institutional owners can hold thousands of vendor agreements across properties and managers. The invoice reaches accounts payable. The signed rate, renewal notice or permitted increase may be sitting in SharePoint, an inbox or a local folder. Getting a payment approved and establishing what a vendor agreed to charge are different assignments.

MARC connects those records, extracts the terms and compares billing with the agreement. For the person calling a vendor about an overcharge, the useful output is evidence: the rate, the clause and the properties covered. The conversation has somewhere concrete to begin.

Work-Bench reports that MARC manages $100M+ in annual vendor spend across 30 US states and Canada, with customers representing about $90B in assets under management. Magnolia Capital and Birge & Held are among them. Those figures describe customer scale and spend under review, with the traction reported by the investor.

In MARC's Birge & Held case study, verified contract coverage rose from 19% to 91% across 24,000 units. Bringing more than 1,800 live agreements into a shared record gave finance and operations a way to connect payment activity with signed terms. The company-hosted account describes using that evidence to secure vendor credit.

That work can travel into the next negotiation. Work-Bench sees a broader opportunity in anonymized, opt-in spending data, with the scope, location and terms needed to make comparisons useful. A landscaping price carries very little meaning until the buyer knows which work came with it.

The same contract can shape a payment dispute, a budget and a decision to renew. Keeping its terms accessible gives those teams a shared starting point.

MARC was profitable before this raise, according to Work-Bench. Its current openings include a Head of Engineering and a GTM Lead / Strategic AE. Aaron Devitt is building capacity to bring the product into more portfolios, where every new property adds another set of agreements for the finance team to understand before renewal season arrives.

Kevel

James Avery wants retailers to own the media business. The advertiser would also appreciate getting home before mastering another buying console.

That customer has a vote in Kevel's expansion. Retailers can control their data, inventory and ad decisions, yet a brand buying across those networks still needs a manageable way to plan, book and measure the work. Independence gets harder to sell when it comes with more administration.

On October 21, the Durham, North Carolina company announced $52M in new capital from Fulcrum Equity Partners, dunnhumby Ventures LLC, Iberis Capital and ORIX USA's Growth Capital business. ORIX confirms its participation. Kevel says cumulative capital now exceeds $100M. AdExchanger identifies the financing as a Series D; the official announcement leaves the instrument breakdown undisclosed.

Founder and CEO James Avery started the business as Adzerk in 2010. CTO Tim Ewald now oversees the cloud platform. The progression is consequential: providing ad-serving APIs has grown into supplying the machinery another company uses to operate a media business, with its own shoppers and commercial priorities attached.

Tesco Media makes that assignment concrete. Kevel powers its new Ad Manager, bringing planning, booking, measurement and optimization into a client-facing platform. Initial placements are live, with further capabilities arriving in phases. Tesco keeps control of customer information and how advertising appears across its channels.

Then there is the problem beyond any single retailer. Dunnhumby's Network Alliance is being built with Kevel to connect campaign activation and measurement across retailers. The June announcement named Tesco, B&Q, John Lewis and Waitrose in pilots. A retailer gets to keep its identity while giving a buyer a less fragmented route to spend.

Kevel already extended its reach through the 2025 acquisition of Nexta, adding off-site advertising and self-service capabilities. Avery told AdExchanger that potential acquisitions of adjacent vendors are on the agenda. No target or signed transaction was announced with this financing.

The new capital will also support AI decisioning, global expansion and advisory services, according to Kevel. Explainable decisions give a retail team something useful to discuss when an advertiser asks why a placement won. Connecting those decisions to a buying process people can actually use keeps the conversation moving toward the next campaign.

BioStack

A medical AI model can get the answer wrong and still collect a passing grade. BioStack's founders have been studying how that happens.

In company-published research, Sanat Mishra and Parth Patwa tested automated graders against medical answers whose meaning had been altered. Some graders continued to approve the answer. That puts a rather consequential question inside the training process: what, exactly, are we rewarding?

Sanat Mishra, co-founder and CEO, and Parth Patwa, co-founder and CTO, are building a business that reaches across that question. BioStack supplies clinical data and training environments for healthcare and drug-discovery AI, bringing patient histories into the exercises models use to improve.

The San Francisco company announced $9.5M in seed funding on October 22. Named backers include Peak XV, Afore Capital, DCVC, Emerson Collective, SV Angel, Verdict Capital, Streamlined Ventures and Y Combinator. The money is intended for US and international expansion, platform development, and engineering and research hiring.

There is a procurement decision hiding in the word environment. The scoring method becomes part of the purchase. An AI lab is buying somebody's choices about which records belong together, what information a model gets before making a decision, and how the response will be judged. A good-looking score can travel a long way before someone asks how it was earned.

BioStack's materials describe patient information running across notes, imaging, labs, medications, treatment and outcomes. Its reported network exceeds 100M de-identified records across the US, India, Brazil and Argentina. That is company-reported reach; the useful unit for a researcher still has to be assembled around a particular question.

The founders bring complementary histories to that assembly. Sanat Mishra worked in cancer genomics and computational biology; Parth Patwa previously worked in generative AI at AWS and research at MIT. BioStack's current openings include a clinical data lead and an RL/reasoning research engineer, putting both sides of the assignment on the hiring plan.

For a model developer, this makes the supplier conversation more demanding. Ask how a case was built, what evidence the grader sees, and whether changing an answer's meaning changes its score. BioStack is raising money to connect the clinical record to that training loop, where the next patient detail and the next grading decision both have work to do.

Maxinsights

The camera can stay on through an entire shift and still miss what a robotics lab wanted to learn. For Maxinsights, catching that mistake early is part of the product.

Founder and CEO Yuguang Yuan and co-founder and CTO Kun Wang have built a business around recording people doing real work, then turning the footage into training data. On October 22, Maxinsights announced a $30M Seed round led by Peak XV Partners, with Tola Capital, South Park Commons, INCE Capital and Alumni Ventures participating.

The Santa Clara company handles collection, checking, annotation, curation and delivery. That puts the person wearing the camera and the lab buying the dataset inside the same operating chain. A customer's request can influence what gets recorded next, instead of leaving the buyer to search through whatever happened to arrive.

MaxLoop checks incoming footage and feeds corrections back into collection. A hand disappearing behind an object can be a routine moment for the person working and a problem for the model trying to understand the movement. Finding it while collection is still happening gives the operation a chance to adjust.

Maxinsights reports more than 2M captured hours, including 1.5M it describes as model-ready. Those are company figures. The distance between those totals is useful to keep in view when a training-data supplier talks about scale: somebody has to pay for the recording, processing and review that happen before an hour becomes something a lab can use.

The next measurement reaches beyond the camera. MaxGlove, being developed with Genesis AI, is designed to capture touch and force. Watching a hand close around an object leaves important information missing about the pressure applied. Adding that measurement changes the recording equipment, the collection process and what the resulting dataset can explain.

There is technical demand behind this work. Dyna Robotics reports human-to-robot transfer improvements as human-video pretraining grows, with the learning objective also affecting the result. For a data buyer, that makes the specification matter: which interactions, which measurements and which coverage the next collection should deliver.

Peak XV is backing a company that has taken responsibility for that whole assignment. As Maxinsights adds force sensing to its collection network, the lab ordering the data gets a more demanding question to answer: what should the next person wearing the equipment teach the robot?

Enact Intelligence

The robot's motor gets a second job at Enact Intelligence: telling the model what the movement felt like. That gives a hardware decision a place inside the AI's education.

Founder and CEO Manuel Yves Galliker has raised $30M in pre-seed funding for the Zurich startup, announced October 22. Notion Capital and redalpine co-led, with firstminute capital participating. The company also names Yann LeCun, Arthur Mensch, Andrew Lonsberry and Thomas Wolf among its angel investors.

Galliker worked at 1X and Physical Intelligence before building Enact. His company is bringing model architectures, data pipelines, actuators and sensors into the same engineering effort. Redalpine describes proprietary actuators designed to make the forces passing through them easier to read.

The attraction is practical. A camera can help locate an object; handling it also involves contact, pressure and resistance. Enact wants that physical information available to the model through the machinery doing the work, alongside visual inputs. The aim is more useful learning from each interaction.

Choosing the body therefore helps choose the training data. Use somebody else's hardware and a model team inherits some of that supplier's decisions about what can be measured. Develop the hardware too, and the team can work on the quality of the observations as well as the intelligence interpreting them.

That is a demanding way to build a company. Hardware changes need engineering time, and the people improving a model have to stay close to the people making the machine. The $30M will support team expansion and faster hardware and R&D development, according to Tech.eu.

Enact is pursuing manipulation work with enough variation to make automation difficult. For a prospective industrial customer, the interesting possibility is a machine that can learn from physical differences instead of requiring the surrounding work to become perfectly predictable.

Performance comparisons and named production customers remain undisclosed in the launch materials reviewed. Enact describes the learning and reliability benefits as its ambitions, leaving industrial buyers to assess the emerging system against their own work.

Enact is hiring in Zurich as it develops that relationship between the machine, its observations and its model. A customer's unusual handling task could eventually become part of what the next robot learns, putting the early deployment conversation inside the engineering roadmap.

Rivian / Third Dimension AI

A rare driving encounter has terrible scheduling habits. Rivian has bought a way for its engineers to call it back.

On October 22, Rivian announced it had acquired Third Dimension AI, bringing 3D scene generation inside its autonomy organization. Financial terms were undisclosed. The useful question is what engineers can do with a road encounter after the vehicle has moved on.

Third Dimension's SuperSim combines reconstruction of real environments with generative simulation. Recreate a situation, vary the conditions, examine how the driving system responds. That gives development teams more control over the examples available for training and evaluation, including situations they cannot conveniently wait to encounter again.

CEO and co-founder Tolga Kart and co-founder and CTO Piotr Sokolski bring experience connecting virtual environments to autonomous systems. The company's 2024 funding announcement described Kart's work at Parallel Domain, Tesla and Activision, and Sokolski's work at Wayve and Google. Their careers have crossed the distance between building a convincing world and using one to develop a machine.

Third Dimension announced a $7M Seed in 2024 led by Felicis, with Abstract Ventures, MVP, Soma Capital and Solari Capital participating. That history explains who backed the company. It supplies no purchase price or investor-return calculation for this acquisition.

Rivian SVP of Autonomy & AI James Philbin says the technology will support both consumer and robotaxi roadmaps. Third Dimension's leadership and technical talent will join Rivian's autonomy research and engineering teams, putting the people building the environments closer to the people deciding what the driving models need to learn.

There is a demanding relationship inside that arrangement. A simulated scene has to be useful for the question an engineer is asking. Third Dimension's own world-models essay discusses how a generator still depends on the diversity of its training data. A convincing reconstruction carries work behind it: choosing the variations, checking their behavior and deciding which findings deserve confidence on a real road.

For a startup whose early pitch reached across gaming, defense and autonomous vehicles, Rivian brings a specific development agenda. The next request for a different driving situation can now travel between colleagues responsible for the simulator and the car learning inside it.

Warp Sciences

If Warp Sciences cannot make the compound, its customer does not pay for it. An ambitious AI lab has accepted a very practical payment condition.

The South San Francisco startup emerged from stealth on October 22 with a $215M Series A led by Lightspeed. Premji Invest, Mayfield, IVP, Lux Capital and Menlo Ventures joined a broader syndicate backing a US chemistry contract research organization, or CRO.

Its customers are biotech and pharma research teams. During lead optimization, scientists need variations of a molecule made so they can compare what those changes do. A model can suggest the next structure. Somebody still has to deliver the material the research team can evaluate.

Warp starts with analog synthesis for preclinical small-molecule programs. The work spans milligram to multi-gram scale, typically 5-15 synthetic steps. The service is priced per compound delivered, with negotiated terms, rather than by full-time-equivalent staffing.

That choice makes reliability personal to the supplier. An unsuccessful synthesis still consumes skilled work, equipment and materials. Under Warp's stated payment rule, Warp absorbs the exposure when the requested compound never arrives.

CEO Kevin Weil and CTO Manohar Paluri bring experience from OpenAI and Meta. CSO Sol Reisberg brings medicinal chemistry experience, including Terremoto Biosciences. COO DJ Patil brings a background spanning the White House and Devoted Health. All four are co-founders, combining the disciplines the laboratory needs in the same company.

Experienced chemists will lead as Warp progressively automates the work. Its founders plan to capture experimental outcomes, including failures, to help train better models and improve the lab. The chemistry service gives that research effort a recurring operational assignment: make something a customer needs.

Warp advertises domestic collaboration at offshore pricing. The public pages describe a small number of design partners; they do not yet supply comparative cost or turnaround results. Customers choose the molecules and retain control of their drug programs. Warp is building the infrastructure around that work, with no drug pipeline of its own.

For Sol Reisberg, who has bought outsourced chemistry himself, the conversation now continues from the supplier's side. Warp's chemists and its design partners work together week by week, deciding what deserves the next attempt and getting the compounds back into the customer's research.

Pythag Tech

Sami Nabulsi wants food companies to license the recipe for making meat from cells. Pythag Tech is taking responsibility for the experiments behind it.

The New York company announced a $4.25M Seed round on October 22, led by Azolla Ventures with Endurance28 and other institutional investors participating. The money expands a business built around selling production know-how to the people who will manufacture the food.

Sami Nabulsi co-founded Pythag with Anne-Sophie Mertgen in 2023. She is its chief R&D officer and previously co-founded Micro Meat. CTO Haytham Samad and chief science officer Tiziano Barberi round out a leadership team spanning computation and cell biology.

That combination has a practical assignment. Growing meat from animal cells involves choices about the cells, the nutrients feeding them and the conditions inside a bioreactor. A promising result still has to become a process a producer can use repeatedly, with economics that support a food business.

Pythag runs experiments across those three areas and uses the biological data to guide its machine learning. The models help choose further experiments, whose results return to the dataset. The laboratory keeps giving the software something worth learning from.

For a food company, the appeal is access to that accumulated research without building the whole discovery operation itself. Pythag's planned product is the production process, including the cells, media and growing conditions. Manufacturing the meat remains the customer's job. Pythag's commercial future therefore depends on somebody else's manufacturing decision.

The spending plans follow that division of work. Sami Nabulsi told Green Queen the company plans to buy automation equipment, expand its wet-lab and AI teams, and open a second laboratory in Thailand alongside its Melbourne lab. More experimental capacity gives the platform more opportunities to learn from actual biology.

Azolla's participation puts climate-oriented capital behind the work of making the production process useful. The disclosed round funds research capacity; Pythag has not published the headline cost or cell-density figures that would let an outside buyer calculate its manufacturing economics.

The next food company considering a license will bring its own product requirements into the conversation. Pythag's growing collection of experiments has to help that producer decide which cells to grow, how to feed them and what to carry into the reactor.

Formic

A 30-person factory can rent a robot. Hiring the department to keep it useful is a different budget conversation.

Formic's customer story about Floorguard Products gets unusually specific about that problem. The coatings maker had no on-site maintenance team. Its sales and operations lead, Jameson Clay, needed somebody to handle implementation and upkeep while his own people kept the orders moving.

Saman Farid, CEO and co-founder, and co-founder Misa Ilkhechi built Formic around taking that assignment. The manufacturer gets the equipment, software, installation, maintenance, financing and performance monitoring through one recurring monthly payment. Expertise arrives with the machine and stays on the account. That is a staffing decision as much as an equipment procurement decision.

On October 22, Formic announced $50M in combined Series B and debt financing. Mercato Partners led the Series B, with Light Ray Ventures, Lightscape Partners, Citi and other investors participating. S2G Investments supplied debt financing. The combined headline belongs to both forms of capital.

The financing supports a move into general-purpose robotics and applications including metal fabrication. Formic says it already operates hundreds of robots across 100+ facilities in 30+ states, with more than 750,000 hours of live production, 99.6% uptime and 97% customer renewals. Those are company-reported figures, attached to an operating fleet with an ongoing service bill.

That last detail matters for the buyer and the investor. A recurring payment is an invitation to keep judging the supplier. Maintenance, parts and deployment speed become commercial responsibilities every month the equipment remains on the floor.

Formic says it is working with Physical Intelligence to adapt robot foundation models for manufacturing. Better models could widen the tasks available to customers, while the installation, support and production-monitoring work still needs somebody's name on it.

For a smaller manufacturer, this changes the first automation conversation. The question can start with a production task and the support needed to keep it running. That gives the plant team a way to add capability without first assembling its own robotics department.

As Formic takes more capable systems into more demanding applications, its service teams inherit those applications too. The next machine arriving at a customer's loading dock brings another job for the people responsible for keeping the line moving.

ZwitterCo

Fat, protein and cleaning chemicals all get a vote in a food plant's water-reuse plan. The production line supplies the ingredients, and the treatment team has to make the mixture behave.

That is the working environment behind Ecolab's strategic investment in ZwitterCo, announced October 22. The amount is undisclosed. The useful detail is who arrives with the money: a partner supplying water-treatment chemistry, digital monitoring and on-site expertise to go with the membrane.

Alex Rappaport, co-founder and CEO, built the Woburn, Massachusetts company with co-founders Christopher Drover, its CTO, and Chris Roy. Founded in 2018, ZwitterCo grew from Tufts-linked membrane research into an industrial filtration business addressing the organics that make water difficult to recover.

A membrane earns its place by separating what a plant wants to keep from what it needs to remove. Organic fouling makes that assignment expensive: material accumulates, performance suffers, and somebody has to clean the equipment. ZwitterCo uses zwitterionic chemistry to resist that buildup, opening more difficult streams to treatment and reuse.

The announced collaboration focuses on food and beverage manufacturing, including brewing, dairy, grain and protein processing. The aim is to recover utility-grade water from wastewater that has been difficult to recycle. That description matters. A plant still has to establish where the recovered water can safely and usefully go.

Ecolab gives that conversation more people with relevant jobs in it. Filtration, chemistry, monitoring and field support have to work together through changes in the stream coming off the production floor. For the operator signing up, the relationship continues long after the equipment order. The plant needs a treatment program that can live with the cleaning schedule, the production mix and the people responsible for both.

ZwitterCo already has venture backing. DCVC led its $33M Series A in September 2022. Its $58.4M Series B in July 2024 was led by Evok Innovations, with DCVC and other investors participating. That round funded an earlier stage of expansion; Ecolab's new investment adds a disclosed strategic relationship without a disclosed price tag.

The company says manufacturers in more than 20 countries use its membranes. The partnership now puts its technology beside Ecolab's treatment programs, where food and beverage customers can evaluate water recovery alongside the daily work of keeping a plant running.

Warhound

Retirement belongs in Warhound's seed-round story. A company proposing to change how working dogs are produced is also taking on the question of who looks after them when the work ends.

Ben Lamm and Palmer Luckey conceived the Texas-based venture, now led by CEO David Spirk. Warhound has emerged with $200M in Seed funding to develop a domestic supply of working dogs for military, law-enforcement and emergency-response teams.

The backers include Sequoia Capital, In-Q-Tel, ARK Invest, Franklin Templeton, 8VC, USIT, Bob Nelsen, Breyer Capital, CAZ Investments, Neogenesis, Builders VC and Lux Capital. That is a substantial group financing an unusually biological version of defense infrastructure.

Warhound plans to combine biotechnology licensed from Colossal Biosciences with biobanking, cloning, training and animal care. Anduril's equipment and software would support protection and situational awareness for the dogs and their handlers.

A dog can have the traits a program wants and still face a demanding journey through training, placement and service. Putting more responsibility inside one supplier gives the buyer somebody to ask about how those stages work together. It also gives that supplier fewer places to leave an awkward handoff.

Spirk's assignment reaches well beyond getting promising science out of a laboratory. A handler needs an animal and equipment that work together. A procurement team needs a dependable source. The animal needs care that continues after its usefulness to the buyer has changed.

Warhound says American Humane Society guidance will inform the program, with planned training certification and support for retirement reunification and adoption. Those commitments belong in the operating design from the beginning. They will also need evidence from the animals' actual lives. A retirement plan written at launch gives that promise a place beside the research agenda.

The financing backs development. Warhound has yet to establish the health improvements, training results or operating economics its proposed program is intended to deliver. Treating those goals as finished achievements would skip the work Spirk has been hired to organize.

For Colossal, this puts licensed biotechnology inside a business serving working-dog buyers. For Anduril, the relationship extends equipment around a living teammate. Warhound has to keep those ambitions connected as a puppy becomes a trainee, a handler's partner and eventually somebody's retired dog.