Humba Ventures is an early stage fund in the Susa Ventures family. Humba invests $250k-$750k into pre-seed and seed companies focused on deep tech and American Dynamism. We invest in areas like defense, robotics, manufacturing, climate and energy, space, and so on.

Susa Ventures leads or co-leads seed rounds with $1m-$2.5m checks. We mostly focus on B2B SaaS, fintech, logistics, and healthcare, and some of the companies we've worked since their early days include Robinhood, Flexport, Andela, Mux, Viz.ai, and Mashgin.

• Designed, implemented, and Hadoopified most of Factual’s data processing pipeline, including the data extraction, data cleaning, and entity resolution frameworks. The entity resolution implementation is a key piece of Factual’s dataset curation process, as well as the basis for Factual’s flagship Crosswalk and Resolve APIs (TechCrunch coverage: http://goo.gl/UC6pK ). • Designed and implemented a distributed in-memory search engine. The search engine indexed 500m documents, ran across 10 machines, and had a continuous uptime of almost one year before being decommissioned for business reasons.• Non-technical accomplishments: did recruiting outreach, which was eventually responsible for 20% of Factual’s engineering hires; was heavily involved in the interview process; wrote company blog posts and answered questions on Quora; helped shape company culture (started annual happiness survey, influenced recruiting process and performance reviews, etc.); co-presented talk at O’Reilly’s Strata Conference (http://youtu.be/pR3WdGbDKiI).

• Designed and implemented most core infrastructure for payment fraud detection. Projects included the signal and model computation frameworks, configuration manager, and automated model definition generator. • Implemented various fraud signals, including tumbling detection and gibberish detection. • Designed and implemented a generic ETL framework to get raw data from external sources (files, databases, logs, etc.), transform that data using the signal computation framework, and save results to the fraud data warehouse. • Rewrote the exception handling system, which reduced the production exception rate by 90%. • Aggressively refactored the codebase, reducing its size by 50 KLOC. • Introduced performance-boosting features like signal evaluation caching across multiple requests, signal definition caching across models, and optimal signal evaluation ordering. • Developed sandboxes for fraud signal and model testing. • Implemented Blogger integration and several new styles for Google Docs. • Non-technical accomplishments: maintained internal wikis; wrote numerous design docs and code tutorials; mentored interns and full-time employees; conducted dozens of interviews.

• Worked on most website features released between late 2003 and late 2005. Wrote code for the DB, business logic, and GUI tiers for projects that included PayPal integration, address book parsing, enabling users to select request routes, LinkedIn Groups, and LinkedIn Jobs. • Designed and implemented a new algorithm for finding paths between members that was faster than the previous algorithm by a factor of over 200. • Tuned performance of graph algorithms and other bottlenecks in the system as LinkedIn grew from 30,000 users to 3,000,000 users. Rewrote several JDK classes to improve performance. • Non-technical accomplishments: mentored new developers; wrote design specs for new features; conducted dozens of interviews.