Avi Patel: Building KLED, Selling Data, Buybacks and More | TG Podcast
Summary
KLED has moved from promise to consumer traction, with 25,000 app downloads in under four days and close to $1 million paid to data contributors. Avi Patel expects one million downloads by year-end. Thread Guy argued that surviving an exceptionally hostile token launch and continuing to ship can ultimately earn a project and its holders credibility: “You survive it, and then you sit down and you still ship… then you’ll get rewarded for it.”
The core business is an enterprise-driven market for legally licensed, highly specific human data—not a flat-price repository of generic camera-roll content. A dancing video might command a premium from Epic Games for AI-generated emotes but only a nominal rate from Microsoft Azor; KLED therefore wants to become “the biggest data collection engine,” then match each dataset to the buyer with the strongest need.
Superpower’s $1 million LOI is KLED’s clearest proof point for this demand model. Avi described the roughly $300 million healthcare company as intending to allocate that amount toward datasets for an in-house healthcare LLM, with KLED proposing anonymized integrations into portals where patients communicate with doctors. Avi’s pitch is that rare consumer-sourced conversations cannot simply be obtained elsewhere.
Avi said a third of relevant enterprise proceeds goes toward token buybacks, but explicitly called current purchases “nominal buy pressure that means nothing” beside actual execution. Roughly 10 million tokens have already been removed from circulation; his hypothetical is that the mechanism becomes consequential at much larger scale—for example, a $1 billion contract would direct $333 million to buybacks.
The durable AI bottleneck, as framed by Thread Guy and endorsed by Avi, is authentic human data because models cannot endlessly train on their own synthetic output without degradation. Thread Guy called this “AI inbreeding”: repeatedly recycling generated golden-retriever images produces progressively less faithful dogs, while sourcing real images may also cost less compute. KLED says its validation stack checks duplicates, internet plagiarism, EXIF metadata and AI generation, reporting about 95% detection effectiveness and aiming for “literal 99% certainty” that delivered data came from a real user’s device.
Avi sees the token less as an enterprise product requirement than as a distribution and fundraising accelerant. Users do not need KLED to use the app, while green candles, referral activity and millions of impressions create measurable FOMO among creators and Silicon Valley investors. His thesis: “Tokenized startups at early stage will always beat untokenized ones,” though he believes crypto needs several genuine success stories before the stigma disappears.
KLED’s next growth loop combines controlled onboarding with gamified data collection rather than opening an economically unsustainable floodgate. Avi is tripling each waitlist batch as contracts grow and previewed KLED Games: timed prompts, potentially backed by a $50,000 prize pool, would ask users to capture specified real-world images while computer vision verifies submissions. Near term, he plans to spend 15 days pursuing larger San Francisco contracts, ship the game, expand PR and pursue centralized-exchange listings.
Deep dive
1. KLED is trying to replace launch-era skepticism with shipped product
Avi introduced KLED as the data-licensing next vertical to Nitrily, his music-licensing marketplace serving hundreds of thousands of artists. KLED lets people license camera-roll and personal data to AI and technology companies, then returns part of the proceeds directly to those contributors.
Thread Guy described Kled as “probably the most flooded token ever in ICM” and said surviving that early period had hardened him. Avi separately said the team kept building through what he called a nightmare: over four months, KLED published roughly 200 updates and delivered an app that would ordinarily represent “a year’s worth of work minimum.”
The launch produced 25,000 downloads in less than four days, against a stated target of one million by year-end. Avi also cited close to $1 million in user payouts, creators backed by around 500 million followers and meetings that week with OpenAI, Anthropic and Google, while saying he had also been signing LOIs with major labs.
The lean operating structure is part of Avi’s explanation for speed: Eli and two other engineers handle engineering, with an AI engineer planned, alongside a design lead, a 3D creative lead, a UGC operation and what Avi called a “fat legal team.”
2. Data becomes valuable when it answers a buyer’s narrow training need
Avi’s framing: there is no universal price for an image or video because value follows the enterprise use case. TikTok-style dancing footage could be premium input for an Epic Games system generating Fortnite emotes, while the same footage might have little special value to Microsoft Azor.
Baseline quality still matters—resolution, useful text, clean images and accurate labels—but scarcity has shifted from having “a lot of data” to possessing specific data unavailable elsewhere. “Everything’s valuable, like literally all of it,” Avi argued, provided KLED can classify it and find the buyer whose model needs it.
The host’s practical question—why not submit his own streams?—produced a concrete rate: Avi said roughly 100GB of high-quality video could earn about $1,000, depending on volume and demand. A well-shot, free-flowing podcast could train conversational or podcasting systems, while location-specific dashcam footage serves a wholly different buyer.
Enterprise acquisition currently comes through investor introductions, online attention and inbound requests from both data owners and data buyers. Avi argued that crypto Twitter’s visibility carries into Silicon Valley Twitter: “The hype online is literally the same thing for Silicon Valley.”
3. Superpower illustrates the case for anonymized, private-domain data
Avi described Superpower as a roughly $300 million healthcare company working toward an AI “super doctor” trained on real exchanges between patients and leading clinicians. The missing ingredient is not public medical prose but data showing how people actually communicate with doctors.
KLED’s proposed role is to integrate with healthcare portals where people already talk to their doctors, anonymize the data and provide it to Superpower. Avi stressed privacy as essential: Superpower would receive useful conversational patterns without knowing the identity behind a record.
The announced $1 million LOI means, in Avi’s explanation, that Superpower would allocate $1 million toward buying datasets; it is an allocation framework rather than a completed purchase. The larger thesis is repeatability: if KLED can prove that a million users will provide requested data, it can build bespoke integrations for many enterprise contracts.
Legality is central to the sales pitch. Avi argued that labs sometimes source data illegally or from wherever they can find it, exposing themselves to enormous future claims; KLED instead aims to offer “the most legally licensed way possible,” making clean provenance valuable both operationally and during fundraising.
4. Contract scale drives both user economics and the token flywheel
Avi said a third of relevant contract proceeds goes into token buybacks. He said approximately 10 million tokens had already left circulation, with execution including Jupiter and Phantom Swap, and emphasized that purchases could occur at random times rather than on a predictable schedule.
His concession was unusually direct: present-day buybacks are “literally meaningless” compared with founders shipping useful products. The mechanism matters at scale, not because a small market buy creates a temporary candle; his hypothetical endpoint was a $1 billion data contract producing $333 million of buybacks.
In a later exchange, Thread Guy referred to Avi putting $33,000 into the chart. Avi said there might eventually be a more effective process, but confirmed that KLED had used Jupiter and Phantom directly.
Asked whether eight- or nine-figure contracts were realistic, Avi said they occur routinely and claimed Whimo had bought “I think $4 billion” of dashcam footage. The host pressed him on the source of that data, and Avi admitted, “I actually don’t know the buyer for this one,” while maintaining that autonomous-driving systems require enormous, area-specific datasets.
His stronger concrete logic was geographic: expanding into Los Angeles requires footage from Los Angeles roads, and KLED says it already has roughly 4,000 Uber drivers there uploading. Avi described KLED’s own contract progression as approximately $50,000, then $350,000, then a $1 million LOI.
5. Authenticity becomes more valuable as synthetic media pollutes training sets
Discussing Sora 2, Avi called it “really, really good” but argued that better compute applied to broad public datasets is only one stage. KLED’s enterprise portal could instead surface extremely narrow material—his example was thousands of gymnastics videos uploaded by a coach—for the next model’s specific weaknesses.
Avi first offered the thesis that labs “slopifying” social media and the internet could increase the value of provably human material. Thread Guy then supplied the “AI inbreeding” explanation: train a model on generated golden-retriever images, feed those outputs back into training, and each generation may drift further from a real dog. He argued that real data may also be more sustainable than repeatedly generating synthetic examples.
KLED says its ingestion system detects duplicates across accounts, searches the internet for plagiarism, inspects EXIF metadata for origin and alteration, and uses Mistral to flag AI-generated files. Avi put current effectiveness at about 95% and the delivered dataset’s target assurance near 99%.
For KLED Games, validation would also be capture-first, “like BeReal,” rather than letting players upload arbitrary saved files. Someone asked to photograph nearby cars would use the live camera, closing the obvious loophole of downloading ten car images from Google.
6. The token is a consumer-acquisition and investor-FOMO mechanism
Users do not need the token to access KLED, although Avi expects future token-locking features. The bigger near-term obstacle is crypto usability: he claimed to have personally walked perhaps 10,000 newcomers through SOL and Phantom, while KLED works toward embedded wallets for non-crypto-native users.
When the host asked whether creators were being paid to promote the app, Avi said the vast majority were friends in his network rather than paid placements; he cited one Instagram Story receiving about 100,000 impressions from an account with roughly 30,000 followers. Future UGC opportunities remain part of those relationships.
Avi said the token’s chart acts as a FOMO benchmark for venture investors. He described investors watching candles, sending messages after price increases and seeking allocation or offering introductions—even though company investors do not receive tokens. He also said the token had become worth slightly more than the company’s current roughly $40 million valuation and referenced a planned $100 million raise.
His broader claim is that KLED accumulated traffic in the millions, 30 million Twitter impressions and rapid app adoption partly because the company and token community grew together from inception. An already-established company launching a token later might receive less incremental benefit: “Tokenized startups at early stage will always beat untokenized ones.”
7. KLED Games and relentless delivery are the next attempt to widen the moat
Avi is throttling waitlist admission because admitting more suppliers than contract revenue can support would underpay contributors. Each batch is being tripled, while referral codes both advance users in line and keep KLED circulating daily across social platforms.
KLED Games would open a parallel path: a hypothetical $50,000 contest sends prompts every 30 minutes—such as capturing 20 local car images—with only seconds to respond. Everyone supplies data, the last surviving player takes the main prize, perhaps 100 others receive smaller rewards, and the frantic real-world participation creates viral clips.
The host’s AI-talent thesis led Avi to say there are “literally no good AI projects on chain” beyond a small handful, call most products vaporware and argue that 99% of ICM projects have become “coming soon farms.” Only repeated delivery, he said, will attract serious traditional AI and Silicon Valley builders back into crypto.
On competition with Launchcoin, Avi denied that its valuation creates KLED’s ceiling and said he no longer wants to farm drama: “Putting out an article with whatever big lab for a billion dollars—that’s engagement farming.” His next 15 days are focused on San Francisco contracts, followed by KLED Games, broader press and outreach for centralized-exchange listings.