Somewhere in the back rooms of biomedical labs, researchers are still squinting at thousands of microscopy images, clicking through analysis pipelines one frame at a time. It's tedious work—the kind that eats up days or weeks of a grad student's life and delays publications.
Alchemy, a two-person venture out of Y Combinator, thinks there's a faster way.
The startup's pitch is straightforward: conversational AI that turns natural-language questions into automated image analysis workflows, all running locally on a lab's own hardware. Upload your microscope images, describe what you're looking for—cytotoxic effects across compound screens, say, or immune cell infiltration patterns—and the system builds a custom pipeline.
According to marketing materials, that process can be up to 98% faster than traditional methods. (The company's own site also cites "90% faster" in some contexts—both figures appearing across their promotional channels.) Either way, the value proposition is clear: compress what once took days into hours, maybe minutes.
Whether that claim holds up across the messy, idiosyncratic world of real lab work is another question entirely.
Plain English, Then Code
The interface itself is deceptively simple. You describe your analysis goal in plain language—no scripting required upfront. The platform interprets that request and assembles a node-based workflow, each processing step laid out visually so researchers can see exactly what's happening under the hood. Click into any node, adjust parameters, swap out algorithms. If something's missing, the system's AI agent can generate new analysis blocks on the fly.
For the more technically inclined, there's an AI-assisted code editor lurking beneath the surface. That dual approach—approachable for bench scientists, flexible enough for bioinformaticians—feels like Alchemy's bid to avoid the usual trap of no-code tools that work beautifully in demos and fall apart the moment you need something custom.
Model training is baked in. The app ships with pre-trained segmentation models for common imaging tasks, but labs can fine-tune those models on their own data. Human corrections loop back automatically, sharpening accuracy over time. That matters especially for specialized assays or imaging modalities where off-the-shelf models struggle—unusual tissue types, rare cell populations, exotic fluorescent markers.
Outputs include reproducible documentation, the kind that could theoretically land in a methods section without major editing. Workflows can be saved, shared with collaborators, or published to a broader community library. In theory.
The Privacy Bet
Here's where Alchemy diverges from much of the competition: the software runs locally.
Image processing happens on the user's machine by default. Research data doesn't leave the lab unless someone explicitly opts into cloud services—limited, per the company's terms of service, to authentication, licensing checks, and optional workflow backups. No images uploaded for model training on remote servers. No centralized data lake harvesting metadata from every institution using the platform.
That's a calculated stance. In biomedical research, regulatory compliance isn't optional. HIPAA, GDPR, institutional review boards—these constraints can derail software adoption faster than any technical shortcoming. By keeping data local, Alchemy sidesteps some of those landmines, at least in principle. The company's website indicates that SOC 2 Type II and HIPAA compliance audits were in progress as of their last update.
Still, the local-first architecture positions Alchemy differently than cloud-based incumbents like Aiforia or Indica Labs' HALO platform, both of which rely on centralized infrastructure for much of their processing power. Whether that trade-off—potentially slower performance on individual workstations in exchange for data sovereignty—will matter to enough customers is an open question.
Who's Behind It
The team is minimal. Phillip Baek, the CEO, studied biomedical engineering at UT Austin and logged time at Amazon, plus research stints at Stanford and back at UT Austin. Liam McBride, the CTO, comes from computer engineering at the University of Illinois, with startup experience and a Stanford affiliation. The company lists San Francisco as its headquarters. Current data suggests a headcount in the 2–10 employee range, though Y Combinator's profile still lists it as a two-person operation—a reflection, perhaps, of how fluid early-stage rosters can be.
Target customers include biotech and pharma researchers, academic imaging cores, and labs running high-throughput assays: organoid screens, wound-healing migration studies, cell tracking in time-lapse videos. There's at least one public endorsement, from Jashwin Sagoo at UC Santa Cruz's Ottemann Lab, crediting the platform with eliminating bottlenecks in microscopy video analysis.
Pricing remains opaque. The website funnels inquiries to a demo request form. Terms of service mention subscriptions, invoices, and potential custom workflow development under statements of work—signals of an enterprise licensing model aimed at institutional buyers, not individual researchers pulling out credit cards.
A Crowded Field
Alchemy isn't walking into an empty room. Established players have been building specialized tools for years: Indica Labs for pathology, Visiopharm for digital slide scanning, ZEISS arivis Pro for large-scale 3D datasets, Oxford Instruments' Imaris for preclinical imaging. Open-source stalwarts like ImageJ, Fiji, CellProfiler, and QuPath anchor countless academic workflows and cost nothing.
Market estimates vary widely depending on who's counting and how boundaries are drawn. Grand View Research puts the broader digital pathology market around $1.5 billion recently, projecting growth toward $3 billion in the coming years. DeciBio's analysis narrows in on roughly $1.2 billion for image analysis tools, with growth rates hovering near 12% annually. Mordor Intelligence slices the image analysis software segment even thinner—closer to $460 million, expanding at just over 9% a year.
Those are moderate growth rates compared to more explosive sectors—solid, but not spectacular. What Alchemy seems to be betting on is that conversational workflows, no-code fine-tuning, and local-first privacy compose a differentiated enough package to carve out early traction. Maybe that's enough. Or maybe it's the kind of feature set that sounds compelling in a pitch but gets lost in the messy reality of institutional procurement, where established vendor relationships and IT department inertia often trump technical elegance.
Open Questions
For now, the company is still in demo mode—finishing compliance audits, collecting feedback from early adopters, figuring out which features actually move the needle. There's a long distance between a polished desktop app and displacing entrenched platforms with decades of domain expertise and customer lock-in.
But the microscopy bottleneck is real. Labs are hungry for tools that let researchers focus on science instead of pixel-wrangling. If Alchemy can deliver on speed without sacrificing accuracy, and if the local-first architecture resonates with compliance-conscious institutions, there might be room to grow.
Whether two founders and a clever interface can scale into a venture-backed business that justifies Y Combinator's backing—that's a different sort of experiment, one that'll take more than a few benchmark tests to resolve.
