Tony Yip

• Designed the full chat workflow, including the core “Chat with Entity (CWE)” system. Built an automated entity-profile knowledge base for cryptocurrencies and U.S. equities, covering project fundamentals, price trends, on-chain and trading metrics, social sentiment, research reports, and deep-dive analyses. • Applied context engineering to address recall challenges in financial-trading scenarios characterized by high asset volume, rapid information updates, and instant query needs. Achieved over 80% recall without relying on generic web search. • Expanded the system to cover 20,000+ crypto projects and the full universe of U.S. stocks, adopted as official search infrastructure by industry media outlets such as BlockBeats and PA News. • Built a multi-path retrieval system for finance and crypto that balances timeliness and domain expertise. For high-timeliness, low-expertise needs, developed an RSS-based breaking-news pipeline with millisecond-level latency and a data platform to support event-driven AI analysis. • For high-timeliness, high-expertise tasks, created a custom timeline-analysis pipeline that structures news, research, and KOL posts into events, enabling event-level semantic search and entity tracking. Users can explore complete historical trajectories and related entities for any project or event. • For low-timeliness, high-expertise retrieval, built a semantic system for research reports and whitepapers, addressing deep-knowledge queries beyond the reach of web search. Developed reusable crawler rules and ingest prompts to maximize retrieval quality. • Led the full design and iterative optimization of a natural-language trading framework enabling users to write strategies in natural language and automatically convert them into JavaScript scripts. • Integrated 200+ financial analysis tools—including trend indicators, social metrics, on-chain analytics, spot/options market data, capital flows, and macro factors—ensuring high composability and interpretability of strategy logic. • Built AI-driven data-visualization capabilities for traders and analysts. Users can generate candlestick charts, volume profiles, capital-flow diagrams, volatility surfaces, sentiment heatmaps, and more through natural-language commands. • Enabled automated plotting, trend analysis, and anomaly detection for both on-chain indicators (TVL, active addresses, gas usage) and traditional metrics (IV, gamma exposure, beta, factor exposures). • Designed a “data-insight chain-of-thought” template that allows models to infer insights from visualizations, such as trend-reversal signals, anomaly explanations, and cross-indicator correlations. • Implemented an end-to-end automated pipeline—from user query to indicator extraction, data retrieval, visualization, and structured conclusions—significantly enhancing non-technical users’ analytical capabilities. • Led the development of a Twitter bot for crypto traders (similar to AskPerplexity), enabling users to trigger automated responses by mentioning the bot under any tweet. The bot averaged ~1M weekly impressions and acquired ~40K new users, becoming a key growth channel. • Led an automated year-in-review campaign built from users’ historical tweets. Core metrics doubled during the campaign, generating over 60,000 sessions. Improved triggering logic, templates, and safety mechanisms (blacklists, rate-limiting), increasing stability and safety. • Introduced feedback and correction pipelines to improve response quality and thematic consistency. • Built a fully automated evaluation system for both offline test sets and live sessions, enabling zero/low-touch error attribution and performance assessment. • Developed a session-analysis toolkit capable of generating traces, locating issues, and categorizing anomalous behaviors in a multi-layer structure. Achieved 100% coverage of all sessions, significantly reducing optimization costs.