Abstract
HomeAdvisor is a conversational chatbot purpose-built for home searching in the Charlottesville, Virginia area. Developed by Alex Song, Rex Wang, and Kyle Chau, the project replaces the traditional filter-heavy interfaces of real estate platforms with a natural language experience powered by a Retrieval-Augmented Generation (RAG) pipeline. The system allows users to type queries such as "Find me 3 bed, 2 bath homes under $500k" or "What's the cheapest home available?" and receive conversational, human-readable responses drawn from a structured database of real listings. The technical stack integrates a Next.js frontend with AWS Bedrock for large language model inference and Supabase as the relational database backend, sourcing listing data from a snapshot of realtor.com via the HomeHarvest Python library. Validation testing combined scripted functional tests of the web UI with exploratory testing of the language model, confirming strong performance across a wide variety of query types while revealing areas for improvement such as emoji input errors, incomplete multi-turn context, and insufficient off-topic guardrails.
The STS research paper examines how the autonomous vehicle (AV) industry deploys engineering-first solutions to deflect questions of corporate liability and public safety. Drawing on Alvin Weinberg's concept of the "technical fix" and Langdon Winner's theory of technological somnambulism, this paper argues that the AV industry's common response to safety failures, including more robust algorithms, better sensor fusion, and improved simulation, creates a black box that excludes non-engineers from meaningful participation in how autonomous vehicles are regulated. Rather than pausing deployment or submitting to independent oversight, corporations like Tesla, Uber, and Waymo treat each fatal incident as an isolated bug to be patched rather than as evidence that the underlying deployment model itself may be flawed.
This paper grounds its argument in concrete case studies, most notably the 2018 death of Elaine Herzberg, in which Uber engineers had deliberately disabled the vehicle's emergency braking system to improve ride smoothness. When the system failed to brake for a pedestrian it had detected six seconds prior, legal consequences fell on the human safety driver rather than the corporation that designed the failure condition. Tesla's marketing of "Full Self-Driving" further illustrates the liability gap: the company sells the experience of autonomy while legally requiring manual oversight, exploiting the well-documented vigilance decrement that makes sustained passive monitoring nearly impossible. Through the Social Construction of Technology framework, this paper demonstrates that the AV industry attempts to force "closure" on the safety debate by monopolizing the language of risk, framing concerns in jargon such as "edge cases" and "neural weights" to disempower public critique.
To address these structural failures, this paper proposes an audit framework built on three pillars: strict liability for black box failures, a clinical trials model for algorithm deployment modeled on the FDA's pharmaceutical approval process, and a legal right to a meaningful non-technical explanation for any automated decision that causes harm. The framework shifts regulatory procedures from "permissionless innovation" to "permissioned deployment," requiring corporations to demonstrate accountability rather than forcing the public to prove negligence.
Both projects engage with the same underlying tension: the gap between what AI systems promise and what they actually deliver, and the question of who is accountable when they fall short. HomeAdvisor demonstrates AI deployed responsibly within narrow, well-defined constraints, while the STS paper examines what happens when that restraint is absent at the level of public adoption. Together, the two projects reflect the same conviction: responsible development of AI-driven systems demands not just better engineering, but transparent accountability structures that keep human judgment and public trust at the center of the conversation.
