ReactWise Cut Pharma Experiments 95% — The How Is Inside

From PhD to Platform

ReactWise, a no‑code AI platform born from PhD research at the University of Cambridge, has cut experimental cycles for pharmaceutical reaction optimization by up to 95 percent — and twelve pilots with pharma companies, including five of the top ten, are underway with conversions to paid subscriptions expected.

Daniel Wigh and Alexander Pomberger completed their PhDs in Chemical Engineering at the University of Cambridge, where their work sat at the intersection of machine learning, chemical engineering, and lab automation. Wigh focused on Bayesian optimization, transfer learning, and data engineering for chemical process development. Pomberger spent five years developing optimization algorithms for pharmaceutical manufacturing. Both had run the experiments, waited for results, and watched the trial‑and‑error loop consume months that a model could collapse into days. The conversation that started ReactWise was not about founding a company. It was about the absence of a no‑code platform that would let a chemist specify a reaction and get optimized conditions without writing Python.

The company incorporated in July 2024. By March 2025 it had secured a place in Y Combinator's Summer 2024 batch and closed a $3.4 million pre‑seed round: a $500,000 Y Combinator check, a £1.2 million Innovate UK grant (about $1.6 million), roughly $1.5 million from unnamed venture capitalists and angels, and a separate UKRI ZeroShotAPI award of £1.18 million for zero‑shot machine learning applied to universal reaction condition parameters. The product launched as an "AI copilot for chemical process optimization", a phrase the founders use deliberately to signal that the software sits beside the chemist, not in place of them. The platform runs a browser‑based UI for designing experiment plates, receiving recommendations, simulating outcomes, and visualizing campaigns, plus an API that connects directly to liquid handlers, flow setups, PAT instruments, OPC UA servers, databases, and electronic lab notebooks. Both paths share ISO 27001 certification and SOC 2 Type II compliance. The business model is annual software licenses; ReactWise does not manufacture robotic hardware.

Before the first pilot, the founders built their own screening laboratory. Starting in August 2024, they ran high‑throughput experiments, 300 reactions at a time, to generate the training data their models would need. By March 2025 they had screened over fifteen thousand process‑relevant reactions, primarily homogeneous chemistry with solid‑handling capability for starting materials. The target: twenty thousand data points covering the most important reaction classes, a dataset they argue is the competitive moat — pretrained models that understand chemistry from day one, so a client can submit a reaction and receive recommendations before running a single experiment. Pomberger told TechCrunch the first full‑scale subscription conversions were expected later in 2025. The timeline from PhD defense to active pilots was less than a year, a pace set by the urgency the founders knew from the lab bench, not from a pitch deck.

How the Engine Works

ReactWise's engine is MemoryBO®, a multi‑task Bayesian optimization (MTBO) algorithm that treats reaction optimization as a sequence of related learning problems rather than a series of isolated searches. Standard Bayesian optimization starts blind: every new reaction begins from scratch, with no memory of what came before. MTBO changes that by jointly modeling reactions that share reagents, mechanisms, or condition spaces, palladium‑catalyzed cross‑couplings such as Buchwald–Hartwig and Suzuki‑type couplings, amide couplings, SNAr chemistry, and learning which historical campaigns are most informative for the current target. Before a single new experiment runs, the model already arrives at an informed starting point.

The transfer‑learning layer is the differentiator. "With transfer learning, we've shown that the number of experiments can be reduced by over 90% — without sacrificing process insight," Pomberger said. In published work, the company reports that optimization beginning from informed conditions instead of a cold start can dramatically cut experimental burden, often by more than 90%. A case study on the company's site shows a target of 15 experiments to hit 95% yield reached in 12. The platform's no‑code interface lets chemists upload prior data from similar or related experiments, map it to the target task, and receive intelligent predictions from day one. An ELN connector pulls internal data directly into the transfer‑learning pipeline.

The approach has caveats. The company's own technical blog notes that transfer learning is "one of the fastest ways to introduce hidden bias into an optimization loop." When reactions are closely related, learning accelerates; when they aren't, the model has no built‑in mechanism to admit uncertainty about that assumption, and performance can quietly degrade. ReactWise mitigates this by letting prior data help early, then fade out as task‑specific measurements accumulate, a more forgiving strategy when prior experiments only partially apply. Chemistry data is inherently noisy: yields can easily differ by 15–20% across labs using different protocols, quantification methods, and calibrations. The guidance is pragmatic — use high‑quality internal data first; if unavailable, use consistent generated datasets; treat literature data as a supplement, not a foundation.

Proprietary data remains the moat. "The high‑quality data we generate and curate is not available to general models, and that is exactly what enables more trustworthy, chemistry‑specific recommendations," the company states. Large language models can already propose decent generic starting conditions, but they fall short on specificity: functional‑group intolerance, complex substrates, and encoding solvent and process descriptors such as polarity, miscibility, and solubility in a disciplined way. ReactWise's bet is that the combination of MTBO, in‑house curated datasets, and a no‑code layer that puts the output in the hands of bench chemists, not just ML engineers, compresses the process‑development timeline from years to months.

What Early Adopters Report

Process development has long been the hidden bottleneck in drug manufacturing. Once a promising molecule emerges from discovery, pharma firms must devise a scalable synthesis route, identifying catalysts, solvents, temperatures, and purification steps that deliver consistent yield and purity at clinical‑trial volumes. The industry has relied on trial‑and‑error experimentation or exhaustive screening, both of which consume months and millions in reagents and labor. ReactWise entered this gap with twelve active pilot programs across pharma companies, and the early data points to a step‑change in efficiency.

Pomberger quantified the macro impact in a March 2025 TechCrunch interview: a typical drug takes 10–12 years from start to launch, with process development consuming 1–2 years. "If we can basically speed up here the workflows — reduce it by an average of 60%, then we can get an idea of how much an effect it is." That 60% average speedup, if borne out across a portfolio, compresses the critical path to clinical supply.

The platform's transfer‑learning layer is a force multiplier. By pre‑training on the startup's own screening library, that dataset, the model enters a new client project with chemical priors rather than a blank slate. Clients also report faster organizational adoption. Pharmaron's testimonial highlights that ReactWise "significantly sped up our adoption of Bayesian Optimization and Machine Learning", a reminder that the no‑code interface matters as much as the math. The Pfizer case study illustrates the multi‑objective reality: maximizing yield while minimizing impurity across a highly interdependent design space, where a single temperature‑profile insight shifted the outcome. The early evidence suggests the bottleneck is widening.

The Market Context

The market for AI‑driven reaction optimization is expanding at a pace that mirrors the broader AI‑in‑chemistry surge. Maximize Market Research values the wider AI in Chemical Market at $1.38 billion in 2025, projecting a climb to $7.85 billion by 2032 — a 28.2% CAGR across the 2026–2032 forecast window. Retrosynthesis planning, an adjacent segment, shows a similar trajectory: $1.2 billion in 2025 rising to $8.7 billion by 2033 at 28.3% CAGR, per Dataintelo.

Innovate UK's Sustainable Medicines Manufacturing Innovation Programme (SMMIP), backed by the 2024 Voluntary Scheme for Branded Medicines Pricing, Access and Growth Investment Programme, deployed £14 million across 29 projects in early 2025. Two competition streams, Early Stage Innovation (EOI) and Collaborative R&D (CR&D), distributed £1.3 million to 15 projects and £13 million to 14 projects respectively. The overall scheme aims to unlock up to £400 million from industry partners across clinical trials, health technology assessment, and manufacturing.

Science Minister Lord Vallance framed the investment as a competitiveness play: "Reducing waste and increasing efficiency will also help ensure British life sciences businesses remain competitive, so this £108 billion sector continues to grow and create jobs." Health Minister Karin Smyth tied it to the government's "Plan for Change," emphasizing high‑skilled jobs and economic growth. Dr Stella Peace, Innovate UK's interim executive chair, called the collaborations "pivotal to improving the way we make medicines." Joe Edwards of the ABPI warned that "in the face of fierce international competition for life science investment, the UK must work hard to stand out from the pack."

Where the Competition Stands

ReactWise enters a field where the battle lines are drawn between drug discovery and drug manufacturing — two adjacent but distinct problems that demand different data strategies. PostEra, founded in 2019, has built its reputation on the discovery side: its machine‑learning platform designs synthetic routes, iteratively refines candidate molecules, and guides experimental testing for biopharma partners. The company reports $26 million in venture funding and more than $1 billion in AI partnerships, including multi‑year agreements with Amgen, Pfizer, and the NIH, according to Crunchbase and PitchBook data. In January 2025, PostEra announced an expansion of its Pfizer collaboration to $610 million, a figure that dwarfs ReactWise's pre‑seed round. But the two startups operate at different stages of the pipeline. PostEra's platform helps chemists decide what to make; ReactWise's no‑code copilot helps them figure out how to make it at scale.

That distinction shapes the data each company prioritizes. PostEra's models learn from literature reactions, patent databases, and partner‑generated data to predict retrosynthetic pathways and reaction outcomes for novel scaffolds. ReactWise, by contrast, has spent the past year running high‑throughput screening in its own Cambridge lab, that throughput, to generate thousands of proprietary data points on reaction parameters like solvent, temperature, and catalyst loading for known transformations. "We are the only ones that have the capability of, and that are currently generating, these high‑quality datasets in house," Pomberger said. "Most of our competitors, they provide the software. The clients are basically prompted with instructions based on the inputs." The claim underscores a fundamental split: discovery platforms lean on public and partner data; process‑optimization platforms argue that manufacturing‑relevant conditions, impurity profiles, yield at scale, robustness, require bespoke experimental data that doesn't exist in public repositories.

Schrödinger, a public company with decades of physics‑based modeling heritage, represents a different competitive vector. Listed among PostEra's top alternatives in 2026 rankings alongside Recursion Pharmaceuticals, Insilico Medicine, Exscientia, and Atomwise, Schrödinger has recently layered machine learning onto its molecular‑simulation engine to accelerate lead optimization. Its approach remains rooted in free‑energy calculations and docking, computationally intensive but physics‑grounded, whereas ReactWise's Bayesian optimization and transfer‑learning framework is explicitly data‑driven and designed for the sparse‑data regime of process development. Other startups like Genesis Molecular AI and Numerion Labs, both venture‑backed, also target early‑stage discovery rather than late‑stage manufacturing.

The market reflects this segmentation. The global AI in chemical market was valued at that level and is projected to reach the 2032 projection with the same CAGR. But the revenue pools are separate: discovery platforms monetize through partnered pipelines and milestone payments; process‑optimization platforms sell subscription software to process‑chemistry teams at pharma and CDMOs. ReactWise's twelve pilot programs with top‑10 pharma companies suggest the latter model is gaining traction, even as PostEra's $610 million Pfizer deal signals that discovery budgets remain larger.

The competitive dynamic may ultimately be complementary. As TechCrunch noted in March 2025, "other startups are applying AI to different aspects of drug development… there's likely to be compounding effects as more automation innovations get folded in." A molecule designed by PostEra's retrosynthesis engine still needs a scalable route; ReactWise's transfer‑learning models, trained on in‑house HTE data, could provide that route with 90% fewer experiments. The winner in each segment will be the platform that owns the data loop — generation, model, deployment, feedback, for its specific problem. ReactWise has bet that loop lives in the manufacturing lab, not the discovery cloud.

Hiring in Cambridge

ReactWise is hiring two engineers in Cambridge, a front‑end engineer and a full‑stack engineer, at £45,000–£60,000 plus equity, hybrid, start ASAP, no visa sponsorship. The postings appear on the company's careers page, Y Combinator's job board, and Handshake. The front‑end role asks for React and TypeScript experience through work, projects, or coursework; a GitHub or portfolio link is required. A bachelor's degree in computer science, engineering, or equivalent practical experience is listed. The company explicitly notes it cannot sponsor visas and welcomes London‑based candidates on hybrid arrangements.

Cambridge supplies the talent pool. The Global Innovation Index 2024 ranked the city the world's most intensive science‑and‑technology cluster for the third consecutive year. Cambridge Consultants, a deep‑tech powerhouse investing ahead of the curve in AI, biotechnology, quantum, and advanced computing, anchors a network of specialists who move between corporate R&D, university spin‑outs, and early‑stage ventures. ReactWise's co‑founders, Wigh and Pomberger, emerged from the university's PhD programme in Chemical Engineering, which cross‑trains chemists and machine‑learning engineers — exactly the hybrid profile the front‑end hire will sit beside.

Innovate UK grants, ReactWise holds a £1.18 million award for its ZeroShotAPI project, further inflate early‑stage headcount budgets. The hiring signal aligns with ReactWise's product roadmap. Pomberger has said the company targets "one‑shot prediction", ideal experiment parameters on the first model call, within two years. Reaching that milestone means the web app must handle richer visualizations, real‑time robotic‑lab telemetry, and multi‑objective trade‑off dashboards without latency that breaks a chemist's workflow. The front‑end hire will ship those features into a user base that already includes Pfizer and On Demand Pharmaceuticals across twelve active pilots.

Cambridge's cluster intensity, the UK AI funding surge, and ReactWise's specific product architecture converge on a single talent need: engineers who can turn probabilistic model outputs into interfaces a bench chemist trusts at 2 a.m. during a process‑development sprint. The same loop that began with two PhDs choosing a tool over a paper now runs through the hands of the engineers ReactWise is hiring today — closing the circle from lab frustration to the medicine cabinet.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.