Upgrade to Pro to get implementation-ready specs for every company, the full report library, and 5 on-demand report requests per month.
If you only have a few minutes to spare, here’s what investors, operators, and founders should know about Raycaster (F24).
Raycaster is an active New York enterprise software company for regulated life-sciences documents. It is unrelated to the Raycast desktop launcher and is not an AI-coding evaluation vendor. Coding-agent references are analogies for long-running document work.[1]
The company launched in 2024 as a scientific research agent for enterprise sales. It searched patents, papers, clinical trials, podcasts, and technical sources for account-specific buying signals.[2] By 2025, Raycaster had shifted from public-data research into internal regulated workflows: cited research, drafting, dependency analysis, staged redlines, and human approval across clinical, CMC, quality, and regulatory documents.
The pivot is the story. Public-data research lost differentiation from horizontal AI products quickly. Company-specific repositories, permissions, schemas, templates, tools, and acceptance tests offered a deeper workflow position. Raycaster remains active and hiring, but its strategic risks include incumbent bundling by Veeva, crowded specialist competition, unverified product accuracy, and a forward-deployed implementation model that may blur software and services.
Gold House identifies Levi Lian and Anthony Humay as cofounders. Lian earned a Stanford degree in Symbolic Systems and worked at Uber, Intuit, and Solvvy, while Humay developed AI, machine-learning, and design experience at Meta, Apple, and Snap after studying computer science at UCLA.[3] NYCEDC also listed both founders in its 2025 fellowship directory, when Raycaster was still described as customer intelligence for enterprise sales.[4]
YC's current founder display duplicates Lian and omits Humay. That interface conflict does not erase the contemporaneous cofounder record, but Humay's present operating role and ownership are unresolved.[1]
The initial insight concerned scientific research in enterprise selling. Raycaster's launch material described a private investigator for technical accounts, locating signals in sources that ordinary lead tools ignored. Humay's launch post likewise positioned patents, academic papers, and clinical-trial data as inputs for qualifying and closing deals.[5]
TechCrunch recognized the specificity of the account research in December 2024.[6] Yet Lian later said differentiation from horizontal AI products lasted roughly one quarter. Raycaster tested public-data research for life-sciences legal and R&D work, then moved toward internal documentation, packaged workflow modules, company context, and embedded evaluations.[7]
The canonical report format asks for two exact founder quotations. The prepared evidence preserves founder claims and interview findings but not verified verbatim quote text suitable for reproduction. No speech is reconstructed. That gap does not obscure the documented product transition, but it limits claims about the precise reasoning and timing behind it.
The current Raycaster reads Word, PDF, Excel, and PowerPoint files, anchors citations to pages and cells, drafts project-level document sets, proposes diff-aware edits, and tracks changes like commits. It maintains a claimed dependency graph across guidance, prior submissions, lab reports, certificates of analysis, batch records, and dossier sections so one scientific change can trigger an impact review.[10]
The product site centers cited research, staged redlines, and approval before merge. It spans protocols, clinical study reports, Modules 2 and 3, SOPs, spreadsheets, guidance, clinical writing, CMC, nonclinical work, quality, labeling, and submissions.[11][12]
Raycaster positions itself above exports from Veeva Vault, SharePoint, and document-management systems rather than replacing systems of record. The architecture uses off-the-shelf frontier models. Lian's stated differentiation comes from workflow design, context engineering, tools, and evaluators, with a portable context layer of repositories, permissions, schemas, templates, tool specifications, plans, and acceptance tests.[7]
This design fits regulated work. A plausible paragraph is insufficient if its source cannot be located, a change conflicts with another section, or a reviewer cannot inspect and reverse it. Raycaster's product thesis is that provenance, dependency tracking, redlines, and acceptance tests are the unit of value, not raw text generation. No independent benchmark verifies retrieval accuracy, drafting quality, impact analysis, or evaluator performance.
Raycaster targets biopharma teams in clinical and medical writing, CMC and technical operations, nonclinical and toxicology, quality, compliance, labeling, and submissions. Current hiring includes a life-sciences GTM role and hourly CMC and clinical workflow specialists, consistent with hands-on enterprise deployment.[1]
No credible reachable-market estimate was found. Drug-development document work is large, but public evidence does not disclose pricing, customer count, contract scope, usage, renewal, or revenue. Named relationships with Agilent, Yokogawa, and Genefab indicate access, not market share.
Veeva is both incumbent and platform risk. Vault already manages regulatory content, workflows, and submissions, while Vault AI agents operate against native data and documents with application-specific safeguards.[13] Yseop offers source-grounded regulatory content, sentence-level traceability, Word workflows, human approval, and Veeva integration.[14] Certara and AlphaLife Sciences also market AI regulatory authoring and compliance-oriented workflows.[15][16]
Raycaster's counter is workflow depth across systems and model providers. Captured diffs, corrections, approvals, tool traces, templates, and acceptance tests could improve evaluations and switching costs. No retention data proves that flywheel yet. If Veeva bundles enough native assistance, an overlay must win on cross-system scope and faster workflow adaptation.
FDA guidance on AI supporting regulatory decisions raises the credibility bar when outputs inform safety, effectiveness, or quality.[17] Public security claims also do not establish that every deployment is validated for GxP or 21 CFR Part 11 use.[18]
Raycaster appears to sell enterprise pilots and software through demos, but pricing and contract structure are private. The verified pre-seed amount is undisclosed. No reliable evidence supports secondary $500,000 claims or Sacra's conflicting funding headline.[8]
Early deployments are intentionally hands-on. Raycaster maps workflows, codifies customer context, runs pilots, and sometimes recruits former CMC, QA, and regulatory-writing specialists as validators before productizing modules.[7] This can accelerate trust and learning, but creates a services-versus-software risk. Revenue, gross margin, implementation capacity, retention, and contract value are unknown.
Every reported a four-person team supporting Agilent, Yokogawa, and Genefab, plus a second product that won an NVIDIA and Nebius competition. Those are partner-reported relationships and awards, not customer-authored outcome evidence.[9]
Current operation is clear. YC lists global enterprise work, open engineering, GTM, and domain-specialist roles, including a founding engineer offer of $100,000 to $160,000 and 2% to 10% equity.[19] The large equity range is consistent with a very early team, not proof of commercial scale.
Raycaster is active, hiring, live, and selling. There is no shutdown, acquisition, decline, or terminal outcome in the evidence. The canonical format requests a named direct quote supporting the outcome, but exact founder quote text is not preserved in the prepared corpus. No quotation is invented.
Public-data research was easy for horizontal AI products to absorb. Lian's account says differentiation lasted about one quarter, prompting a move toward internal documents, packaged modules, company context, and evaluations.[7] The pivot reduced exposure to generic research competition and increased dependence on customer-specific integration.
Forward deployment helps Raycaster map real work and encode acceptance tests. It also consumes expert capacity and can make each implementation unique. The central risk is whether reusable modules accumulate faster than bespoke requirements. Hiring domain specialists as validators is evidence that trust requires more than model quality.
Raycaster can sit above Veeva and other repositories, but incumbents control permissions, documents, workflow state, and customer relationships. An overlay must prove that cross-system dependency analysis and portable context outweigh the simplicity of native AI. The alternative explanation is that regulated teams prefer independent review layers precisely because no single system contains the whole dossier. Public adoption evidence cannot yet decide between those paths.
Ready to rebuild Raycaster?
Implementation-ready specs, every report, and 5 on-demand requests each month.