15:[[47518119,{"by":"mliezun","descendants":2,"id":47518119,"kids":[47519466],"score":9,"text":"Hi HN!

I'm Miguel, an engineer at Gencove.

We've just launched a new consumer product that provides genetic insights using low-pass sequencing rather than traditional microarrays.

For now we offer 9 polygenic risk scores based on the eMERGE network paper:\nhttps://www.nature.com/articles/s41591-024-02796-z

After receiving results, you can download your VCF directly (no need to go through support).

Happy to answer questions about the tech, accuracy tradeoffs, or how this compares to array-based approaches.","time":1774450347,"title":"Show HN: Gencove low-pass sequencing genetic test with raw data access","type":"story","url":"https://www.gencove.com"}],[45164234,{"by":"rh121","descendants":58,"id":45164234,"kids":[45166006,45164326,45169235,45166692,45168039,45166914,45164682,45165437,45168276,45174538,45169005,45165715,45164489,45165569,45164626,45165668,45164957,45166953,45168324,45165927,45166609,45165922,45168673,45164247,45164932,45170163,45166201,45169254,45167298],"score":23,"text":"Billions of dollars spent, incredible hype that we will have AGI in several years. Does anyone think the current deep learning / neural net based AI approach will eventually hit a dead end and not be able to deliver its promises? If yes, why?

I realize this question is somewhat loosely defined. No doubt the current approach will continue to improve and yield results so it might not be easy to define "dead end".

In the spirit of things, I want to see whether some people think the current direction is wrong and won't get us to the final destination.","time":1757299448,"title":"Does anyone think the current AI approach will hit a dead end?","type":"story"}],[43985732,{"by":"andrew_zhong","descendants":0,"id":43985732,"kids":[43992025,43990132],"score":14,"text":"While working with LLMs for structured web data extraction, we saw issues with invalid JSON and broken links in the output. This led me to build a library focused on robust extraction and enrichment:

- Clean HTML conversion: transforms HTML into LLM-friendly markdown with an option to extract just the main content\n- LLM structured output: Uses Gemini 2.5 flash or GPT-4o mini to balance accuracy and cost. Can also also use custom prompt\n- JSON sanitization: If the LLM structured output fails or doesn't fully match your schema, a sanitization process attempts to recover and fix the data, especially useful for deeply nested objects and arrays\n- URL validation: all extracted URLs are validated - handling relative URLs, removing invalid ones, and repairing markdown-escaped links

I'd love to hear if anyone else has experimented with LLMs for data extraction or if you have any questions about this approach!","time":1747236578,"title":"Show HN: Robust LLM Extractor for HTML/Markdown in TypeScript","type":"story","url":"https://github.com/lightfeed/lightfeed-extract"}],[40394124,{"by":"andrew_zhong","descendants":3,"id":40394124,"kids":[40394202,40395281],"score":5,"text":"We created LightFeed to transform any websites into lightweight and focused news feed. Use LLM to sort and summarize posts using user's custom prompt. For example, you can get anything related to "Games, science or open source" on HN.

The news feed is automated daily and you can receive it in browser, email or RSS.

We will also open source the LLM web parser/sort library soon o (https://github.com/lightfeed/lightfeed). It reads HTML, turns main content into markdown, uses LLM to parse it into structured feed in JSON, then sort with embedding on user query. It is written in Typescript, uses llama index framework, and supports most LLMs on the market.

We are also shaping an enterprise plan that comes with 1000+ news sources already indexed and a real-time LLM query engine.

Any feedback is welcome!","time":1715979215,"title":"Show HN: LLM-powered News Hub, e.g \"Games, science or open source\" on HN","type":"story","url":"https://www.lightfeed.ai"}]]

user:	raphaeltm
created:	February 20, 2017
karma:	2