Participated · Series A · Aug 2026

Westlake Robotics develops unified whole-body large-motion models (GAE) and a dual pretraining architecture that separates a "big brain" (perception and high-level decision) and a "little brain" (millisecond-scale whole-body motion execution). It has released the self-developed humanoid robot body Westlake o1 (2026) and built an end-to-end stack linking algorithms, models, and hardware to create a closed loop of hardware testing, data feedback, and model iteration. The company prioritizes real-world data collection by deploying robots in practical scenarios to gather high-quality whole-body posture and interaction data for training. Its research team includes academics and industry alumni from top universities and companies, with over 300 papers published at conferences such as ICML, NeurIPS, CVPR, and RSS. Financially, Westlake Robotics raised 500 million yuan across four rounds within six months, including a recently completed Series A, to fund R&D and talent development in embodied intelligence.

Led · Seed · Jun 2026

Grace Investment Machine develops financial time-series large models and positions itself as a next-generation self-evolving intelligent-agent investment platform. The company aims to use a multi-agent architecture to create AGI-era investment machines and an asset management platform, seeking to establish a new AI-driven asset management category. It is based in Beijing. Following its recent consecutive angel and angel+ rounds, the company plans to allocate capital to model R&D, build out compute infrastructure, and expand its team. The articles do not disclose revenue, user metrics, or a valuation.

Led · Series A · Apr 2026

Kepler Robotics focuses on industrial-grade, general-purpose humanoid robots and pursues a full-stack in-house development strategy with roughly 80% self-sufficiency in core components. Its flagship K2 "Bumblebee" humanoid is designed for heavy-load operation, extended endurance, and lower deployment costs and is targeted at smart manufacturing, warehousing, special operations, and high-altitude work. In April the company deepened its strategic focus on an "embodied intelligence brain" and force-tactile full-stack data acquisition to address scarcity of physical interaction data, tactile perception gaps, and model generalization. Kepler is building a high-quality industrial physical interaction dataset of tens of millions of entries and develops the VTLA full-perception model alongside its KeplerBrain system. The company has secured benchmark projects in high-altitude live welding, automated production line inspection, and heavy material handling. Kepler is expanding its industrial deployment through its Jiangyin subsidiary and the Xiake Bay Science City settlement, with R&D and marketing teams already onsite.

Participated · Seed · Jan 2026

Theorem is a San Francisco-based startup from Y Combinator’s Spring 2025 batch that automates the formal verification of software produced by AI coding assistants. Its platform blends classical formal methods with proprietary AI models trained to automatically generate and check mathematical proofs, shrinking work that once took PhD-level engineers years down to days or weeks. Central to the product is a “fractional proof decomposition” approach that allocates verification effort according to the criticality of each code component, enabling scalable oversight of large, complex systems. The team believes the next bottleneck in software development will be trust, not code generation, as tools from GitHub, Amazon and Google already create billions of lines of code annually. Early collaborations are under way with customers in AI research labs, electronic design automation, and GPU-accelerated computing, and the company recently uncovered a bug that evaded Anthropic’s internal tests. The $6 million seed financing provides capital to expand the product’s reach and address a widening “supervision gap” threatening critical infrastructure. No revenue or user numbers have been disclosed.

Participated · Series E · Sep 2020

Mohalla Tech is the parent entity of the vernacular social media platform ShareChat and the short-video entertainment app Moj. The company raised an additional $16 million in debt financing from Singapore-based EDBI. That injection brings the total amount raised in its ongoing financing round to $65 million. The financing reported in the article is structured as debt. The report also states the company laid off approximately 5% of its staff. ShareChat is an Indian social media startup focused on live-streaming and has grown in part through acquiring rival MxTakaTak. The company recently secured $49 million in debt financing from existing investors to fund bonus ESOP grants and boost employee morale. Its valuation plunged from nearly $5 billion to below $2 billion after a new funding round. Management has implemented aggressive cost-cutting measures even as the company reports revenue growth. ShareChat faces challenges monetizing a user base with limited purchasing power and carries a high expense-to-revenue ratio. The firm's ability to adapt—particularly around live-streaming—will be pivotal to its competitive prospects in the Indian market. ShareChat, under parent Mohalla Tech, operates social and short-video platforms and includes apps Moj and the recently acquired MX TakaTak. The company reported 180 million monthly active users for ShareChat and a combined user base of 300 million across Moj and MX TakaTak. It has previously raised capital, including a $266 million funding round that valued it at $3.7 billion and included investors such as Alkeon Capital and Temasek. In the current deal Mohalla Tech secured nearly $300 million in fresh funding from Alphabet's Google, the Times Group and Temasek, valuing the firm at nearly $5 billion. The investment is set to be announced soon and marks Google's second major bet in India's short-video space after backing Josh. ShareChat also counts Twitter and Snap among its investors. ShareChat is a Bangalore-based social network that offers its app in 15 Indian languages and runs Moj, a short-video app launched shortly after India banned TikTok. The company claims more than 160 million users and a 50 million+ creator community; Moj reportedly logs 4.5 billion views per day and average user sessions of about 34 minutes. Moj is described by management as the firm's fastest-growing product and is expected to potentially become bigger than ShareChat. ShareChat has raised over $911 million to date and recently closed an additional $145 million tranche as part of its Series F, bringing its valuation to nearly $3 billion (TechCrunch headline lists $2.88 billion). The business competes in short-form video with players such as MX TakaTak, Glance’s Roposo and DailyHunt’s Josh. Management says it will continue to invest in AI capabilities, scale its global AI organization, build advanced editing tools and help creators monetize on the platform. ShareChat is a Bangalore-based social network that offers apps in 15 Indian languages and targets users in smaller cities and towns across India. The company claims more than 160 million users and has expanded its product suite to include Moj, a short-form video app with over 80 million monthly active users. Moj was built rapidly after TikTok’s exit and integrates Snap’s Camera Kit to strengthen its short‑video offering. ShareChat is experimenting with monetization through commerce interactions in audio chat rooms and launched a fantasy sports app called Jeet. The company says it aims to grow its ecosystem of products toward 1 billion+ cumulative MAUs as internet penetration deepens in India. To date ShareChat has raised about $765 million in funding.