Developer maps four AI APIs' incompatible citation formats after weekend of data wrangling

·1 views

A developer running a monthly study on AI-generated source citations found that OpenAI, Anthropic, Gemini, and Perplexity each return cited URLs in structurally different formats with inconsistent naming conventions. Extracting comparable data required writing a separate adapter for each provider to normalize outputs into a uniform structure. A key pitfall discovered was that Gemini returns Google redirect URLs rather than publisher URLs, making all citations falsely appear to originate from a single domain unless resolved via HTTP HEAD requests. Further normalization is also needed to deduplicate URL variants of the same page that differ by protocol, trailing slash, query parameters, or capitalization. The developer recommends building sanity checks for domain-diversity collapses, which typically signal extraction bugs rather than genuine findings.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Kontigo, a YC S24 startup, is hiring a Founding Engineer

Kontigo, a startup that participated in Y Combinator's Summer 2024 batch, has posted an open position for a Founding Engineer. The job listing was shared on Hacker News, Y Combinator's associated tech community platform. The role is listed directly on Y Combinator's company directory under Kontigo's profile. No further details about the company's focus or compensation were provided in the listing.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

How to Merge Data from Multiple Excel Files Using VBA, Power Query, and AI

Data professionals frequently struggle to consolidate specific sheets from large numbers of Excel files into a single unified table. The task grows more complex when sheet names vary across files and data structures are inconsistent, with mismatched headers, column orders, and data types. Manual copy-paste methods are error-prone and unscalable, while VBA macros offer automation but require coding knowledge. Power Query provides a more accessible middle ground for structured consolidation workflows. Modern AI-assisted tools are also emerging as potential solutions to handle inconsistencies and reduce manual intervention in data merging tasks.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Build a WhatsApp Landing Page Clients Can Edit Without Breaking It

A developer tutorial on DEV Community outlines how freelancers and studios can create maintainable WhatsApp landing page templates for clients. The guide recommends defining the page's purpose and editable elements — such as headline, services, colors, and phone number — before writing any code. It emphasizes centralizing the WhatsApp number and pre-filled message in a single, clearly commented constant to minimize handover errors. Placeholder text like [Business Name] and [Service Area] should be used throughout demos instead of fabricated testimonials or metrics. The author also provides a pre-launch checklist and links to a public CodePen demo, with paid licensing options starting at USD 20 per project.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Vaniq-Edge: 34MB Open-Weight TTS Engine Achieves 2.94% Word Error Rate

A developer has released Vaniq-Edge, a lightweight, end-to-end text-to-speech engine with approximately 8.5 million parameters and a file size of just 34.3 MB. The model converts text directly to 24kHz mono audio without relying on external vocoders or secondary models. It achieves a Word Error Rate of 2.94% and a predicted audio quality score of 4.35 out of 5, while running faster than real-time on both GPU and single-core CPU. The model is available as open weights on Hugging Face and can be tested locally using Python. Key limitations include support for English only, a single fixed voice, and occasional high-frequency audio artifacts on certain consonants.

0 comments Read more at DEV Community