PandaAI bets LLM agents democratize quant research; 100k users, three rounds

PandaAI is a Chinese AI trading-research company founded around 2024 by Li Yuqi, a Columbia-trained quant who had been partner at a private fund managing more than ¥1 billion. Its thesis is that large language models will erase the research gap between retail traders and institutional quant teams, so the company sells agentic research infrastructure rather than trading signals or promises of returns.

The founding bet came from two observations. Li believed quant funds' edge was a narrative built on proprietary data, low-latency channels and large research teams — all things LLMs would deflate — and that Cursor proved better tools alone do not make better researchers. His answer is a ladder from Qube (one sentence becomes an inspectable factor-and-backtest workflow) to EVO (an AI-native professional workbench) to OS, an Agent-to-Agent environment where agents do factor iteration and portfolio research under human-set risk rules.

By August 2026 the company had 70+ staff, 100,000+ users inside and outside China, and 20,000 registered users on its overseas version TQX, concentrated in Southeast Asia and the Middle East. It shipped a live futures client in June 2026, open-sourced its first QuantSkills library and agent clusters, and published research on its CQ2 model and A2A self-evolution architecture. Three factor competitions drew more than 30,000 participants, including winners who were not professional quants — evidence for the founder's claim that AI narrows the judgment gap.

The commercial path is B2B first — brokerages, futures firms and funds — to build experience and data corpus, then consumer. Funding came in three rounds totaling tens of millions of RMB, announced August 2026 with angels led by L2F光源创业者基金. The unresolved risk is trust: social media still paints AI trading as a scam category, and Li himself frames the product's limit as human judgment — agents research, people decide and hold risk.