Hongshan's Seed Fund Backs New Startup Building an AI-Powered LinkedIn

"Can AI Connect People to Each Other?" Text by Jiaxiang Shi

Edited by Zhiyan Chen

Anywaves has learned that AI startup Index Gravity completed its pre-A funding round earlier this year, co-led by Hongshan and Alpha Startups.

Index Gravity founder Beichuan Yu was an early core member of Douyin, witnessing its growth from tens of millions of users to 600 million daily active users. He was responsible for building Douyin's early social features.

In early 2022, Yu chose to start his own company. It began as an overseas e-commerce business. Yu sold press-on nails, vacuum cleaners, and even $20 drones. In May 2023, their TikTok shop was shut down, causing a cash flow crisis. They had to stop paying all team members' salaries while scrambling for monetization opportunities — at one point even selling courses on traffic-driven e-commerce — until founding AI marketing company Index Gravity in 2024.

By early this year, before Manus ignited the Agent market, Yu posted a document on WeChat Moments to recruit people, reflecting on his entrepreneurial journey and mobilizing former ByteDance and Douyin colleagues to help spread the word. Unexpectedly, this post attracted Hongshan. The firm recognized their future scalability, and the new funding round came together naturally.

Today, Index Gravity operates in the global short-video creator marketing space, providing a marketing digitization platform for domestic and international sellers. The company has established partnerships with multiple AI startups, with monthly revenue currently around several hundred thousand US dollars.

One could call this a typical AI marketing company, but Yu told Anywaves that their next goal is "AI-powered LinkedIn."

"In the past, almost all AI tools have been in 'single-player mode' — helping you write copy, generate images, draft reports, write code, completing one person's tasks. But the next phase of AI, similar to the evolution of the internet, will definitely connect people to each other," Yu said.

Just before Index Gravity's new product launch (the name remains under wraps for a few more days), we spoke with Yu. He told us he hopes to build a new $100 billion company in this new era. "After all, aim for the moon, and even if you miss, you'll land among the stars."

The following conversation has been edited and condensed by Anywaves —

Anywaves: The leap from AI marketing to AI-powered LinkedIn seems quite large. How did you think about this?

Beichuan Yu: It looks like a big leap, but for us it's the same path. We started with AI influencer marketing because it's the most concentrated, most urgent "find people" scenario: brands need to find creators to collaborate with, demand is clear, frequency is high, and the pain point runs deep.

As we worked, we realized that fundamentally we're not solving "creator marketing" but "finding the right people" — whether that's creators, clients, experts, or partners. So we generalized our capabilities, enabling AI to handle all business relationship scenarios.

So from our internal perspective, this is a very natural evolution — following one problem to a more general solution.

Anywaves: Specifically, what is "AI-powered LinkedIn"?

Beichuan Yu: The AI-powered LinkedIn in our vision isn't a better contact list, but transforming the "relationship network" into a real-time operating intelligent system. Traditional LinkedIn is static — you see profiles, connections, and have to search, add people, and message them yourself.

Anywaves: Who do you envision as the initial users?

Beichuan Yu: **The people we serve first are those who need to "find people" every day. Marketing managers who need to find KOLs, creators, and partners for campaigns; entrepreneurs and BD professionals looking for potential clients, investors, and channel partners; researchers seeking industry experts, speakers, and interview subjects; recruiters searching for candidates and freelancers.

They share one thing in common: spending hours daily searching, filtering, building spreadsheets, and writing outreach on LinkedIn, in databases, and across social media.

Anywaves: You started building this product in June, but the Agent narrative took off in March. What were you doing during those three months?

Beichuan Yu: **Our AI marketing business has been growing consistently these past few months — about $200,000 in revenue last month alone. Making a bold, aggressive pivot from a business with positive feedback to something new is, frankly, a difficult decision. Plus, we kept thinking about what the moat is for vertical Agents.

Anywaves: Did you find an answer in those three months?

Beichuan Yu: **If every company has to bet on one thing, our bet is on "where the boundary lies between large models and vertical applications." I believe the boundary sits at capabilities that cannot be internalized by large models.

Large models improve through massive amounts of public data. But some data simply doesn't exist in the public world. Like proprietary data in specialized industries — forecasting Nongfu Spring's upcoming sales, for instance... Additionally, some products get better through user feedback. Large models can access all publicly available news, but they can't know what you like to read on Toutiao.

The core of Agent products is making every user your model annotator, telling the model what's good and what's bad, enabling rapid feedback and learning. This is something all AI applications need to consider this year.

Anywaves: That sounds like the data flywheel logic again. Kimi pushed this narrative during its massive marketing spend, and it was ultimately proven unworkable.

Beichuan Yu: **This is somewhat different from the model wave. A model's foundational capabilities are determined by corpus quality. The data used in past training sets the model's intelligence at factory release. Today we're more like building containers — data feedback can form closed loops. I naturally know which people you found to be suitable matches.

Anywaves: In the longer term, how will this product evolve?

Beichuan Yu: **If we use AI to reach out to people extensively, then those people will also need AI to handle incoming requests on their behalf. When everyone has an Agent, our interactions become increasingly efficient. Assuming both Agents are built by the same company or platform, that would certainly be the most cost-effective solution.

It's somewhat similar to the bilateral effects of the past internet, but transformed into bilateral agents.

Anywaves: Do you think you have the barriers and capabilities to become a platform?

Beichuan Yu: **We're still very early-stage — we wouldn't claim to have a moat yet. But we're moving in the right direction:

First, polish the experience for single-side users; second, as more people use it, we accumulate increasingly rich people-finding and connection data, making matches more precise; third, once both sides have Agents, we can enable smarter coordination between them, minimizing communication costs.

A moat isn't built overnight — it grows gradually alongside users, data, and network effects. This is a marathon.

Anywaves: "From everyone having their own AI to everyone having their own Agent" — a lot of people are telling this story. Why are you all so convinced?

Beichuan Yu: **Fundamentally, we see AI as a productivity revolution that helps you escape things you shouldn't have to handle. Humans should ultimately just receive results. Beyond close personal relationships and offline meetings, everything else should be optimized by AI and Agents.

Anywaves: Then why do Agent products today have such high failure rates?

Beichuan Yu: **The most likely reason for failure is simply giving it a 30-word prompt without enough context. It doesn't know your underlying needs, can't extend your intent, and when forced to supplement limited text with context, it fills gaps with hallucinations.

Additionally, models lack cognition about tool usage — things can get pretty erratic at first.

Anywaves: So compared to your expectations, the current product performance isn't that good?

Beichuan Yu: **Definitely not good enough yet.

Today, nearly all tool-type Agent launches score 80 on head scenarios but 60 on long-tail needs. For us, our 80-point scenarios are ones where we've pre-trained on data, but other untrained scenarios may depend on model luck.

Anywaves: Then why launch now?

Beichuan Yu: **Users today input in chatbot style — you'll always have unsolvable long-tail needs. You need to launch first, acquire a certain scale of users, get data feedback, to know what's not working well.

Most Agent products can't be excellent at launch, but you need to ensure certain scenarios are good enough, then rely on post-launch team iteration to improve the rest.

Anywaves: In what domains are you confident?

Beichuan Yu: **Professional fields and influencer social media perform relatively well. For example, you can use it to find CMOs at US companies with revenue under $50 million. But finding academic scholars, or who published what papers in which databases — we probably don't do well enough there yet, since we haven't optimized the data for that.

Anywaves: You worked at Douyin for four years. Did Douyin influence you significantly?

Beichuan Yu: **It was a process of demystification. I joined Douyin in early 2018 when the core team had only about a dozen people. Some projects went directly to Yiming and Nan Zhang. Back then, ByteDance believed in young people — I'd only graduated three months when I became responsible for the first CCTV Spring Festival Gala sponsorship red envelope, distributing 1.2 billion RMB with a 60-person project team.

Why demystification? Because it gives you an "illusion" that you're close to them, so you dare to do things and dare to become such a person yourself. This is an era dividend. Those who can capture such dividends are lucky, and there aren't that many of them.

Anywaves: If you had anticipated Douyin e-commerce and TikTok as massive opportunities later, would you still have left?

Beichuan Yu: **I was going to start a company, so I definitely would have left, and the earlier the better. But I left early enough already — couldn't have left any earlier (laughs).

Anywaves: Many founders say entrepreneurship is a process of self-discovery. Do you feel similarly?

Beichuan Yu: On day one of entrepreneurship, you need to recognize you're an idiot. Fresh out of Douyin, I was an extremely浮躁, sharp-edged young person. At Douyin, you unconsciously think you're a little genius, failing to realize these things were given to you by the environment — ego gets huge.

At ByteDance, Yiming liked to say, "All success factors are constructible." That's not wrong, but you can't ignore the difficulty of constructing those factors. Entrepreneurship is a process of continuously leveraging your external resources and adding to them.

Now, I've also recognized that I shouldn't do BD-resource-driven things, but product-and-technology-driven things.

Image source | IC Photo

Recommended Reading

The Flow of Money, The Rise and Fall of People