CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

ainativedev/aidevcon-2026-nyc

AI Native DevCon 2026 New York — practical agent skills grounded in conference talks

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

outline.mdskills/talk-pratik-patel-refined-rag-data/references/

Source outline

Source: Pratik Patel, Data Isn't the New Oil — It's Refined Data, AI Native DevCon New York 2026. Recording. All anchors refer to original numbered cues in the transcript. Short quotations below preserve the supplied transcription, including spoken wording.

TopicSource anchorExact short quoteApplication and scope
Bounded docs demo00:06:18–00:09:17, cues 101–150“an ask the docs assistant.”The example compares raw and refined documentation retrieval.
Stale and duplicate corpus00:09:19–00:10:43, cues 151–172“70% of the chunks come from older releases”Corpus percentages describe the demo, not all RAG data.
Separate useful content00:10:47–00:12:29, cues 173–198“strip away all that noise.”Remove irrelevant page material while preserving meaning.
Rejection ledger00:12:32–00:13:04, cues 199–208“We rejected this data for this reason”The pipeline records excluded content and reasons.
Data linting00:13:08–00:13:53, cues 209–219“our linter.”Domain checks detect obsolete or unsuitable data.
Curated blend00:13:55–00:15:04, cues 220–236“a curated FAQ”Selected enrichment can help retrieval after purification.
Delivery discipline00:17:46–00:21:28, cues 273–330“deliver data like we deliver code.”Mutable work and versioned releases serve different roles.
Incremental storage00:21:33–00:23:36, cues 332–360“only uploads or downloads. Exactly what's changed?”The talk describes incremental artifact transfer.
Tagged evaluations00:23:44–00:25:21, cues 362–388“things that must not be included.”Tagged cases distinguish stale guidance and new features.
Threshold and history00:25:58–00:28:03, cues 397–426“whatever number you decide to choose,”Release criteria are choices and version history supports comparison.
Smaller model hypothesis00:28:58–00:30:51, cues 439–468“You can use a much smaller model,”The demonstration motivates testing a simpler model.
Historical usefulness00:30:55–00:33:09, cues 470–500“migration notes”Older documentation can still carry useful scoped context.

Source limitations

  • The demo uses a narrow documentation task; its corpus percentages, chunk counts, and sample answers do not establish general accuracy.
  • The spoken version and model details are inconsistent or approximate; do not infer exact release histories, model size, or current commands.
  • The source mentions both 90% and 95% gates. Treat them as example thresholds, not a single validated standard.
  • Claims about infrastructure, free tiers, FineWeb comparisons, and storage performance are talk-time descriptions, not current guarantees or independently verified benchmarks.
  • Promotional interlude cues 237–247 are omitted from the bundled transcript; the remaining cue numbers retain their source positions.
  • This is a spoken transcript; unseen slides, linked repositories, and third-party research were not independently inspected. No outcome beyond the described talk is established.

skills

talk-pratik-patel-refined-rag-data

tile.json