数据科学家
Data Scientist
Termgrid 正在重新定义私人资本市场的规则。我们为交易专业人士打造了该领域的标杆操作系统——一个全球最复杂的私募股权发行人、贷款人和顾问管理其融资流程每个阶段的平台。
由 Dipish Rai（哈佛MBA，IIT，哥伦比亚大学；前 Providence Equity）和 Vishal Singh（哥伦比亚大学计算机科学硕士；前 Link3D 首席技术官，后被 Nasdaq: MTLS 收购）于 2019 年创立的 Termgrid，已实现许多金融科技公司梦寐以求的目标：4 家全球排名前五的私募股权公司将我们视为其核心科技合作伙伴。1600+ 机构，30,000+ 专业人士。
职位描述
我们正在为平台添加智能推荐功能：在合适的时机，从社交信号（谁与谁连接，网络活动）和交易信号（谁与谁交易）中，悄然呈现最相关的联系人、交易对手和机会。推荐应微妙且真正有用，而不是销售式的推送。
你将是第一位从零开始构建这一功能的数据科学家，使用大多数公司都渴望拥有的数据。这是一个需要交付成果的“自建自用”角色。
职责
- 构建并负责首个推荐功能，在产品内呈现相关联系人、交易对手和机会。
- 熟练应用经典的推荐技术：协同过滤、聚类、最近邻以及标准的机器学习模型，处理社交和交易数据。
- 从简单开始，逐步引入嵌入、相似性搜索和排序，仅在必要时加入。
- 处理稀疏数据和新用户及实体的冷启动问题。
- 保持推荐内容微妙、相关且可信：在金融产品中提供有帮助的发现，而非销售话术。这既是建模问题，也是产品和用户体验的判断。
- 负责评估：包括离线指标和在线实验，与真实参与度和采用率挂钩，而非虚荣数字。
- 与后端工程团队合作进行服务部署，并在此领域作为首位数据科学家建立轻量、实用的机器学习实践。
要求
- 在生产环境中拥有构建推荐或个性化系统方面的扎实、实际经验。
- 对经典的推荐方法有深入理解：协同过滤、聚类、最近邻以及标准的机器学习模型。
- 推荐系统基础：嵌入、相似性或向量搜索、排序、特征工程以及严格的评估。
- 精通 Python（pandas、numpy、scikit-learn）和 SQL。
查看英文原文
Termgrid is rewriting the rules of private capital markets. We built the category-defining operating system for deal professionals — the platform where the world's most sophisticated private equity sponsors, lenders, and advisors manage every stage of their financing workflows.
Founded in 2019 by Dipish Rai (Harvard MBA, IIT, Columbia; ex-Providence Equity) and Vishal Singh (Columbia MS CS; ex-CTO Link3D, acquired by Nasdaq: MTLS), Termgrid has achieved what few fintech companies dream of: 4 of the top 5 global private equity firms trust us as their core technology partner. 1600+ institutions. 30,000+ professionals.
The RoleWe are adding intelligent recommendations to the platform: quietly surfacing the most relevant connections, counterparties, and opportunities to users at the right moment, drawn from social signals (who is connected to whom, network activity) and transaction signals (who transacts with whom). The bar is subtle and genuinely useful discovery, not a salesy push.
You will be the first data scientist building this, from the ground up, on data most companies would love to have. It is a build-and-own role for someone who ships.
Responsibilities
- Build and own the first recommendation features, surfacing relevant connections, counterparties, and opportunities inside the product.
- Apply classical recommendation techniques well: collaborative filtering, clustering, nearest neighbors, and standard ML models, on social and transaction data.
- Start simple, then layer in embeddings, similarity search, and ranking where they earn their place.
- Handle sparse data and cold-start for new users and entities.
- Keep recommendations subtle, relevant, and trustworthy: helpful discovery in a finance product, not a sales pitch. This is a product and UX judgment call as much as a modeling one.
- Own evaluation: offline metrics plus online experiments tied to real engagement and adoption, not vanity numbers.
- Partner with backend engineering on serving, and set light, useful ML practices as the first data scientist in this area.
Requirements
- Solid, hands-on experience building recommendation or personalization systems in production.
- Strong grasp of classical recommendation methods: collaborative filtering, clustering, nearest neighbors, and standard ML models.
- Recommendation foundations: embeddings and similarity or vector search, ranking, feature engineering, and rigorous evaluation.
- Strong Python (pandas, numpy, scikit-learn) and strong SQL.
- Experimentation literacy: you design and read A/B tests and know the difference between offline and online lift.
- Product and business judgment, plus pragmatism: you can make recommendations feel subtle and relevant rather than salesy, you start simple, ship, and improve.
- Nice to have: capital markets, private credit, or fintech domain exposure; graph or network methods (graph embeddings, GNNs, link prediction) as a bonus for future sophistication; early or founding data scientist experience.
Originally posted on Himalayas