MangoDesk
YC S25San Francisco, US · Founded 2025 · 2 employees · Hiring · 2 known investors
MangoDesk provides reinforcement learning environments designed for training AI systems on real-world business tasks and workflows.
Also known as MangoDesk (YC S25)
Founders & leadership· Y Combinator alumni (S25)
MangoDesk was founded in 2025 by Ankith Subramanya and Ananth Subramanya.
Investors · 2
Company profile
researched Aug 2026MangoDesk is a San Francisco-based company, founded in 2025, that builds data and evaluation infrastructure for AI development. Its stated focus is long-horizon, production-grade reinforcement learning environments intended to make AI systems more capable at real-world workflows, alongside tooling for custom evaluations and post-training data.
The product lets an AI team describe an evaluation or annotation requirement in natural language, after which the platform automatically generates the corresponding annotation pipeline - not only the labeling interface but also annotator instructions and quality-assurance systems - which can then be revised through further natural-language edits. Teams can route work to their own annotators or to MangoDesk's vetted pool of experts, and export the finished dataset. The company frames the alternative approaches as building the whole apparatus in-house, which it describes as slow and interface-heavy, or engaging large data-labeling vendors, which it associates with long turnaround times and limited transparency.
The founding team is two brothers: Ankith Subramanya, CEO, previously a software engineer at Scale AI working on its generative-AI data engine for frontier labs; and Ananth Subramanya, CTO, who previously founded Lumio, a software firm serving Fortune 500 and enterprise clients. Recruiting materials describe the team as drawing on experience from Scale AI and Uber and position the company against the knowledge-work economy it sizes at $5 trillion.
Founding story
MangoDesk was started in 2025 by brothers Ankith and Ananth Subramanya, who present the company as a response to the difficulty of producing custom evaluations and post-training data: building the capability internally requires weeks of interface work, instruction writing, annotator sourcing and QA wiring, while outsourcing to large labeling vendors involves cost, slow turnaround and limited control. Their premise is that specifying evals and data work should be as simple as describing the requirement in natural language. Ankith had been a software engineer at Scale AI on its generative-AI data engine serving frontier labs; Ananth had founded Lumio, a software firm with seven-figure revenue serving Fortune 500 and enterprise clients.
Business model
MangoDesk sells data and evaluation infrastructure to AI teams. Customers describe their eval or annotation needs, generate a pipeline on the platform, and either supply their own annotators or draw on MangoDesk's vetted expert pool, downloading the resulting dataset when complete. Third-party listings characterize the commercial model as SaaS/subscription and enterprise software, though the company's own materials focus on platform capability and a demo/contact-sales motion rather than published pricing.
Third-party sources describe subscription-based software revenue, with one aggregator citing a $29 per month entry price and estimating annual revenue in the range of roughly $0.3 million to $0.6 million; these figures are external estimates and not company-disclosed. The company's own materials describe engagements built around generating annotation pipelines and optionally supplying annotators.
Traction
The company reports a team size of two and, in YC hiring materials, an oversubscribed seed round led by top-tier venture firms. External estimates place annual revenue at roughly $0.3-0.6 million and monthly website visits at about 700; such figures are modeled by third-party providers rather than disclosed by the company. Active hiring in San Francisco covered a founding engineer role at $100,000-$200,000 base with 0.50%-1.50% equity, and later a strategic projects lead.
Latest developments
Following the Summer 2025 YC batch and August 2025 launch, recruiting materials state the company closed an oversubscribed seed round led by top-tier venture firms. Third-party aggregation places a seed round in October 2025 at approximately $500,000 and lists Rebel Fund, Script Capital and Y Combinator among investors, though these figures are estimates rather than company statements. As of mid-2026 the company was advertising two open on-site San Francisco roles, a strategic projects lead in operations and a founding engineer in engineering.
▸Full profile — market position, technology, go-to-market, geography, history, risks & controversies
Market position
MangoDesk is an early-stage, YC-backed entrant in the AI evaluation and human-data infrastructure market, competing with established data-labeling and annotation providers including Scale AI, Labelbox, Surge AI, Sama and 2027.dev. It differentiates on automated, natural-language pipeline generation and on RL environments rather than scale of labeling operations, and remains very small - a team of roughly two people with third-party revenue estimates in the low hundreds of thousands of dollars.
MangoDesk's stated points of difference are specification of annotation pipelines in natural language and automatic generation of the full pipeline - interface, annotator instructions, routing and QA - which can subsequently be edited in plain English, rather than being custom-built. It also offers a choice between customer-supplied annotators and its own vetted expert pool activated on demand, and positions its work around production-grade, long-horizon RL environments tied to measurable model improvement. Its founders' prior experience building generative-AI data labeling infrastructure at Scale AI and enterprise AI systems is presented as part of that positioning.
Technology
The platform converts natural-language descriptions of eval or data requirements into complete annotation pipelines, automatically producing the labeling interface, annotator instructions, routing and QA systems, all of which remain editable through natural language. It is positioned as data and evaluation infrastructure supporting production-grade, long-horizon reinforcement learning environments for real-world workflows. Engineering roles emphasize production-grade backend and full-stack systems, scalability and system design, with applied AI experience in post-training and RL preferred.
Go-to-market
The company reaches customers directly through founder-led outreach, a public YC launch post and demo video, a founders@ email address and a booking link for 30-minute demos. Recruiting and launch copy target AI labs and application-layer companies building evals, training models or running annotation operations. Y Combinator affiliation and job boards serve as additional visibility and hiring channels.
AI labs and application-layer AI companies, specifically teams creating custom evaluations, training or post-training their own models, and teams running data annotation operations.
Geography
Operations are concentrated in San Francisco, California, which is the company's headquarters and the location of its listed on-site roles.
History
The company was founded in 2025 in San Francisco by brothers Ankith and Ananth Subramanya and joined Y Combinator's Summer 2025 batch, where Gustaf Alstromer is listed as its primary partner. It published its YC launch post in August 2025 under the framing "Cursor for Evals and Human Data." Job materials posted on YC state the team had recently closed an oversubscribed seed round led by what it describes as top-tier venture firms. As of the YC listings, team size was two, with hiring underway for a founding engineer and, later, a strategic projects lead.
Risks & controversies
As a two-person company founded in 2025 with limited disclosed financials, MangoDesk faces concentration risk in its founding team and competes with substantially larger data-labeling incumbents, including Scale AI, where its CEO previously worked. Publicly available figures on revenue, valuation and funding come largely from third-party estimators and are inconsistent with one another; one aggregator page describes the company's business as an end-to-end AI recruitment platform, which conflicts with the company's own descriptions of its product.
Compiled by commissioned research from 7 cited public sources — announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed — not independently audited.
Competitors · 2
by search overlap
Companies competing with MangoDesk for the same Google search keywords, organic and paid, via search-intersection analysis.
Timeline · 2
launches, deals, and filingsPublic launch post describing a platform that generates bespoke annotation pipelines - interface, annotator instructions and QA - from natural-language descriptions, with the option to use the company's vetted annotator pool.
MangoDesk was part of Y Combinator's Summer 2025 batch, with Gustaf Alstromer as primary partner.
Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.
In the news
▸Research sources · 7
primary sources listed
- MangoDesk: Long-horizon RL environments | Y Combinatorycombinator.com · web
7 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does MangoDesk do?
- MangoDesk builds long-horizon RL environments and natural-language-generated annotation pipelines for AI evals and post-training data.
- Who founded MangoDesk?
- MangoDesk was founded by Ankith Subramanya, Ananth Subramanya in 2025.
- Who are MangoDesk's investors?
- MangoDesk's investors include Y Combinator, Script Capital.
- Where is MangoDesk headquartered?
- MangoDesk is headquartered in San Francisco, US.