Perle

The frontier AI labs building the next generation of large language and multimodal models face a structural problem that more compute cannot solve: when the easily scraped web data runs out, model quality stops improving, or gets worse. The technical term is model collapse — the recursive degradation that sets in when a model trains on outputs produced by earlier models. The fix is human signal, but sourcing it at scale, from people who actually know the subject matter, has proven one of the more intractable challenges in the industry.

Perle is a Solana-based platform built to address that problem. It connects AI labs and enterprise teams with a vetted network of domain experts, manages the annotation and evaluation workflows, and uses Solana's blockchain to record every data point with an immutable, auditable provenance trail.

The Network

Perle's contributor base spans 49,000 verified experts across 70 countries and 76 languages. The roster includes 2,500 physicians, 800 attorneys, and specialists across two dozen additional fields. All contributors are on payroll rather than treated as gig workers — a distinction the company makes explicitly on its website.

This is the core differentiator Perle pitches to AI labs: not crowd annotation, but expertise-in-the-loop. The difference matters for tasks that automated systems handle poorly. Benchmarks are easy to pass with surface-level pattern matching; cases that require genuine domain judgment — interpreting a medical image, assessing whether a legal argument holds, identifying bias in a clinical trial question — require people who have spent years in those fields.

Perle claims its Expert-in-the-Loop approach outperforms automated tagging systems by 70% on benchmarks, based on comparisons run against Amazon Rekognition. The company has also published research under its "Notes from the Field" series, including evaluations of frontier model performance on Arabic cultural knowledge, code-switching speech, and acoustic sound identification — areas where its expert network provides the ground-truth labeling.

Three Products

Perle organizes its platform into three verticals.

Data Services is the core offering: end-to-end programs covering data acquisition, labeling, model evaluation, red-teaming, and bias auditing. This is where the company works with hospital systems, law firms, and AI labs to produce task-specific datasets and run expert-led safety reviews.

Winnow is a hiring product that uses AI-powered interviews and domain rubrics to help organizations source and vet specialist contributors for their own internal AI pipelines. Rather than outsourcing the annotation to Perle, a client using Winnow is building its own team with Perle's vetting infrastructure.

WhisperMind targets robotics. It captures multimodal streams — RGB-D video, IMU data, and force-torque sensor readings — labeled by contributors performing the physical tasks being taught. The goal is embodied data: annotations produced by humans who have actually cooked the meal, opened the stuck drawer, or picked up the wet glass, rather than someone tagging video footage from a desk.

Solana as the Settlement and Provenance Layer

Perle integrates Solana through a four-layer protocol architecture. The lower layers organize data into workflows and route tasks to contributors based on reputation scoring. The third layer — the Settlement and Record Layer — is where Solana comes in. Every completed annotation is recorded on-chain, creating an immutable proof-of-work log for each data point. This gives enterprise clients a verifiable audit trail: they can demonstrate, cryptographically, where a training example came from and who verified it.

The Solana integration also replaces what has historically been a slow payment mechanism for data workers. Traditional annotation platforms can take 30 to 90 days to process contributor payments. On-chain settlement via Solana brings that down to sub-second finality. Contributors keep 80–90% of their earnings according to Perle's documentation, compared to the take rates common on traditional annotation marketplaces.

Perle's expansion into Solana was formally announced in 2025. Rishin Sharma, Head of AI at the Solana Foundation, was quoted in the announcement: "Perle's approach to scaling human intelligence with a permissionless platform strongly aligns with our vision."

Funding and Backers

Perle raised a $9 million seed round led by Framework Ventures in 2025, bringing total funding to $17.5 million across its pre-seed and seed stages. Earlier investors include CoinFund, Protagonist, HashKey Capital, and Peer VC. The company was founded in early 2024 by AI veterans with backgrounds at Scale AI, Meta, MIT, and Berkeley. CEO Ahmed Rashad has been the named executive across company announcements.

The PRL Token

Perle launched its native $PRL token on March 25, 2026. The token serves several roles within the protocol: it is the payment mechanism for contributor rewards, a platform access credential for priority entry into what Perle calls Expert Guilds, and a governance asset operated through the Perle Foundation, described as a nonprofit steward of the network.

The token has a hard cap of 1 billion units. At the Token Generation Event, 17.5% of the supply entered circulation. The community allocation is 37.5% of total supply, with 7.5% released immediately at TGE. Investor allocations (27.66% of supply) are subject to a 12-month cliff followed by 36-month linear vesting. Team tokens (17%) follow the same vesting structure. The Perle Foundation controls the ecosystem allocation at 17.84%, with 10% unlocked at TGE.

Context and Position

The AI training data market is crowded — Scale AI, Appen, and several newer entrants compete on similar ground — but Perle's thesis is narrower than general-purpose annotation. It is explicitly targeting the hard cases: tasks that require genuine expertise rather than volume, in domains where automated quality checks cannot catch the failures that matter. The Solana layer adds a property no traditional data vendor offers — verifiable, on-chain provenance — which may matter increasingly as regulators and enterprise procurement teams ask harder questions about training data lineage.

The product suite has also expanded since founding. The original framing was squarely blockchain-centric. The current public positioning emphasizes the expert network and the three product lines, with the on-chain layer treated as infrastructure rather than the headline feature. That shift reflects a broader pattern in the Web3 industry: the most durable applications often present the blockchain as plumbing, not pitch.