About
Draper
Draper Associates backs companies, in the firm’s own words, “before the world catches up” to what they’re building. Tim Draper invested in Tesla when electric vehicles were niche, in SpaceX when private spaceflight sounded absurd, and in Coinbase when crypto was perceived as a fad; he also wrote early checks to Robinhood, Twitch, Hotmail, and Baidu. Venture is the family trade: Tim is a third-generation VC, and his children run funds of their own. Today, Draper manages $2.3 billion across more than 400 companies and 60-plus unicorns, investing in AI and robotics, crypto, healthcare, aerospace, and ideas the firm files under “the unconventional.” It still writes some of the earliest checks a founder will ever take. Fund 8, a $300 million early-stage vehicle, launched in 2026.
A firm with that history could coast on reputation, but Draper is doing the opposite. “We’re an extremely AI and data-driven firm,” says Michelle Kwok, a Principal there, and Draper is building the tools to back that claim up. The mandate from the top is expansive: Tim wants the team to answer every founder who applies and to uncover the founders who haven’t raised, or told anyone what they’re building, yet.
Reaching every founder and turning the firm’s own knowledge and judgment into working software both require infrastructure. Draper built a strong MVP and centralized AI-driven database in-house, then realized they were learning about new AI tools from their MBA intern classes, and decided to hand the reins over to them. Now, much of the new-age AI infrastructure is developed by the Master's intern classes, with key guidance from the investment team. The program hands new talent open problems and the resources to solve them, and this year’s class shipped production tools the team now uses every day. Each of those tools runs on data, and much of that data comes from Harmonic’s API.
Open problems, and the resources to solve them
“What sets Draper apart is how much faith they put in junior talent,” explains Akash Khanikor, one of this year’s interns who is building an AI-powered Daily Briefing Tool for the investment team. “We came in as interns, got handed open problems and a lot of AI tools, and we weren’t micromanaged. If we hit a wall, the team stepped in fast.”
Draper’s bet on early talent is the same bet it makes on founders, and it turns on a pitch. In their interview, every intern in this class pitched a tool they thought the firm could use, the same way a founder pitches a company, and were given the go-ahead and the room to build. By Draper’s account, every proposed tool was on point.
What the class shipped reads like a product roadmap: Draper Brain, an assistant that answers any question about any company; the Draper Deal Engine, which scores potential investments against four decades of Draper decisions; automated daily briefings; portfolio trackers. Draper Brain and the Deal Engine both lean directly on Harmonic’s API. Both grew out of the same discovery: the more the interns pushed each tool to cover companies Draper hadn’t engaged with, the more it needed data from outside the firm’s own systems.
Draper Brain: institutional memory, extended by Harmonic
Before the interns built anything, Draper’s own knowledge was scattered across files no one could query. “When I joined as an intern, our company data lived in eight Excel sheets, and people were dropping them into AI to try to surface information,” recalls Michelle, who ran the first pass at centralizing things. She pulled the sheets into a single database, one record per company, then handed the work to the next intern class. Centralized, the records were still only as deep as what Draper had entered by hand. Companies don’t hold still: a round closes, headcount jumps, a new CTO starts, a new product ships, and the database is already behind.
Patrick Blumenthal took the handoff and built Draper Brain, an MCP server that pulls Draper's internal sources, Airtable, Google Drive, and email, into Claude. It now connects 15,000+ contacts, 450+ portfolio companies, 4,900+ co-investors, and 1,900+ meeting notes, and is used daily by the full investment team. Wiring them together turned a tooling problem into a data problem. “I found out fast that so much of the internal data was missing or inaccurate,” Patrick says. “We looked for ways to fill the holes, and then we saw Harmonic, where there’s so much fresh data available.”
Harmonic is the layer that fills what the internal records can’t. Its API opens onto a continuously refreshed database of more than 35 million companies and 195 million people, so a question about a company Draper has never engaged comes back with a full profile. In a live pull, that profile spans headcount and its growth, total raised, current and lead investors, team breakdown, founder backgrounds, and traction signals like web traffic and social following, most of it drawn from Harmonic and merged with whatever Draper already has on file. Harmonic resolves each company and the people around it into a single current record, so Draper Brain answers from live market data instead of a static internal snapshot. Ask for a competitive landscape around SpaceX and the answer stitches three layers together: the pipeline from Draper’s internal Airtable, the portfolio companies Draper has backed or companies in the pipeline, and the wider market from Harmonic.
Draper Brain opens that merged view to the whole team, Tim included. “Instead of wildly searching the web, we get filtered data from Harmonic alongside internal data our companies certify,” Michelle says. “Tim can ask Claude anything about any company, and it searches the internal records, then Harmonic.” The payoff shows up most in diligence. “When we invest, we write a memo, and mapping the competitive landscape used to take half a day of research,” Patrick says. “With Harmonic powering Draper Brain, it takes minutes.”
The Draper Deal Engine: Draper’s judgment + Harmonic’s data
Answering questions about a company is one problem. Deciding which companies earn a memo at all is a harder one, and it’s what the Deal Engine was built for. Draper wanted a model that sizes up a deal the way its partners would, trained on the firm’s own history.
Jiatao Fan and Spandan Pattanayak built the engine. “The original idea was to use both the companies Draper has backed and the ones it’s passed on to build an AI that can evaluate future deals fast,” he explains. The model required two kinds of data Draper couldn’t supply on its own: clean, consistent records of companies from those past decisions to learn from, and a steady feed of new companies to evaluate. Harmonic supplies both. It returns every opportunity in the same shape, which is what a model needs to weigh thousands of them against each other, and each week it adds net-new companies to the queue.
The engine scores each company across eight dimensions, among them contrarian theses, founder credibility, growth potential, and past traction, then rolls them into a single model of how Draper decides. As such, its judgment runs across far more companies than a team could review by hand. Out of the roughly 20,000 companies Draper could plausibly talk to, the model surfaces the 150 or so that fit the firm, so partners spend their judgment on the ones that matter.
At the weekly cadence, the filter earns its keep. Harmonic surfaces a couple hundred companies a week, the model takes the first pass, and only the ones that clear it reach a human. “People want the frontier, weekly. They don’t want two hundred a week,” Michelle says. “The first week Jiatao ran it, two startups passed our model, and I’m happy to look at those two. If we had to look at many times more than that, it becomes less useful.”
Scout: a second opinion on every deal
The Deal Engine’s scores still need checking, and Jiatao runs them through Scout, Harmonic’s AI research agent. Ask it about a company in plain language, and Scout does the research legwork, pulling funding, competitors, founders, and market position from Harmonic’s data and the web into one brief. “Scout is really ChatGPT plus PitchBook,” Jiatao explains. “Whenever the Deal Engine flags a company, I check it against Scout myself. I can immediately see past investments, growth record, the founder’s story, and more, to confirm the model got them right.”
Two things make Scout the tool he reaches for. “It has a more thorough database than the foundational models,” he says, and it connects companies the way an analyst would. “You can pull up any company and ask for its closest peers, or ask for other companies its founders have started, and get answers fast.” A foundation model alone can’t match that depth of coverage, or trace those kinds of connections between companies.
Widening the funnel
The Deal Engine narrows the bottom of the funnel. Harmonic widens the top, surfacing founders before they’re easy to find, like the ones still in stealth, or those who left a job last month and haven’t announced what’s next.
“We don’t want to miss anyone,” Michelle explains; “for example, we want to have access to all the stealth founders.” AI has multiplied the number of companies forming, and “the top of the funnel is larger than it’s ever been.” A field that large has outgrown the personal network Michelle came up sourcing through. “Harmonic opens us up to more possibilities than I’d have had ten years ago. Back then I’d have invested in people I already knew, or whomever my VC and founder friends sent me. Now we see as many founders as possible.”
Data-driven, without the data team
Draper’s claim to be an “extremely AI and data-driven firm” now has tools behind it: Draper Brain and the Deal Engine, in production and built on Harmonic’s API. For a firm with four decades of history, moving at the frontier meant trusting its newest people to build the proof. They did, without a data team of the firm’s own, because Harmonic supplied the element that typically takes years to build: current, structured coverage of the market.
