Abstract
Summary of AutoBleedr
Mountain biking has a global community of enthusiasts who demand the best performance from every component on their bikes. Hydraulic brakes provide superior control and responsiveness that mountain bikers need for challenging terrain, yet they require regular maintenance to remove air bubbles from the brake fluid. These air pockets compromise braking performance because unlike incompressible brake fluid, air compresses under pressure and reduces hydraulic force.
The AutoBleedr targets recreational mountain bikers who want to maintain their hydraulic brakes without the complexity and time demands of traditional bleeding methods. The primary audience includes all cyclists with limited mechanical experience, and bike shop owners seeking to streamline routine maintenance services.
The deliverable consists of an automated brake bleeding system featuring a specialized syringe holder that mounts directly onto the bicycle. The onboard microcontroller initiates and manages the bleeding sequence, controlling the syringe/brake handle actuators and vibration mechanisms that replicate the manual tapping technique used by professional mechanics. By combining microcontroller logic with mechanical hardware, we transform a tedious manual process into an efficient automated procedure.
The tool utilizes stepper motor-driven syringe plungers for consistent fluid manipulation, sensors for real-time bubble detection, and small actuators for brake lever operation. The design accommodates standard brake lever positions without requiring specialized mounting adjustments or positioning procedures. This approach provides the strongest value for SRAM’s full-flush routines where repeatable cycles and sensor-guided purges eliminate manual effort while preserving mechanical reliability.
Abstract of STS Paper
Patent prosecution is a high-stakes translation system that converts invention disclosures into claims and specifications that the United States Patent and Trademark Office (USPTO) can evaluate. Because small changes in claim language can expand, narrow, or weaken protection, inventors often rely on patent agents and attorneys to secure durable protection. As practitioners begin testing generative AI (especially large language models) for summarization, drafting, and review, two concerns stand out: exposure of confidential invention materials and trade secrets, and hallucinated information in filings.
Using Actor-Network Theory (ANT), I analyze AI-assisted patent prosecution as a sociotechnical network in which inventors, disclosure materials, patent agents/attorneys, firm document systems, prompt templates, AI models, USPTO examiners, and court rulings all shape the filing process. I map the prosecution workflow–intake, first draft, review, and filing–and show how risk enters when confidentiality guardrails and verification are weak or missing. I combine document analysis of governance texts and technical risk frameworks centered on federal statutes, USPTO guidance, USPTO Rules of Professional Conduct, federal case law, and ABA ethics guidance, using Florida state bar opinions to show how those duties are applied at the state level. I also use the NIST AI Risk Management Framework to identify where controls are weakest in practice. AI can play only a limited role in patent prosecution when the network preserves accountability, enforceable confidentiality, and human verification of model outputs. Even then, some uses remain difficult to defend, especially when public models receive client disclosures or model-generated claim language and citations survive into the filing.
Relation of AutoBleedr and STS Project
The AutoBleedr and my STS Project, Clients, Counsel, and the Model: An ANT Study of AI in Patent Prosecution, are not closely related. The AutoBleedr is a custom designed tool for automating repetitive and tedious steps for performing brake maintenance on high-end road and mountain bikes. In contrast, my STS project focuses on how artificial intelligence (AI) can safely and effectively be used to aid in patent prosecution, by establishing boundaries. AI is useful in many situations to increase efficiency and help build upon internal prowess. While developing the AutoBleedr, AI was useful to brainstorm and debug technical roadblocks. In patent prosecution however, word choice and validity are high-stakes necessities, limiting the ways that AI can effectively be used.
