Nora23
Nora23 on Hacker News
How does this compare to GPT version in terms of interactive capabilities?
How does this handle images with different compression formats?
How long did it take them to plan this escape?
Does this mean Langfuse will now have better ClickHouse integration?
Good catch! Yeah, it's about ~20 seconds right now with Flash 3.0. Definitely on the slower side.
I'm hoping to move to Flash Lite when it's out, that should help a lot. Also thinking about adding some better loading feedback so it doesn't feel quite so long while you wait.
Posted by
Nora23 on Friday January 16, 2026 @06:17AM
from the show-hn dept.
https://pavo.studio-hedera.com/
Hi HN,I built Pavo Travel, an AI-powered travel guide that generates custom audio tours on-demand for any location.The problem: Traditional audio guides are pre-recorded and limited to major attractions. Generic AI guides suffer from training data cutoff—they don't know about places that opened last year or current hours.Pavo Travel solves this by combining Google Places API with Gemini's...
Same here. The metainterpretation stuff is fascinating but dense.
Makes sense for dev boxes you spin up weekly but don't actually use 24/7. Like staging environments.
By the time GPT 5.5 landed we were already on 5.1, honestly they seem to converge on similar limitations around compositional reasoning.
Fair point! I was hoping maybe there'd be some rumors or speculation floating around, but you're right. Thanks anyway!
Posted by
Nora23 on Thursday January 15, 2026 @06:40PM
from the ask-hn dept.
I'm currently using Gemini 3.0 via API, but it's slow and expensive. On the other hand, 2.5 Flash doesn't have enough capability for what I need.I was hoping for a 3.0 Flash Lite version that balances speed, cost, and performance, but I haven't seen any rumors or announcements. Does anyone know if Google is planning to release one?
Thanks! That's a really smart approach, showing the assumptions upfront makes it transparent.
context switching between tools is painful. card-based triage sounds smart.
been relying on it for 25 years now. hope it sticks around.
nice resource. been using this for years when building chess engines.
pre-commit hooks catch the easy stuff. code review is still useful for architecture and logic though.
The natural language input is a nice touch. I always found traditional calorie trackers tedious with all the database searching. How does it handle portion sizes? Like if I say "had a burger" vs "had a big burger"?
PAVO uses location-based grounding with Gemini to generate local guides in your native language. So it's less about "travel utility tools" and more focused on deep local knowledge and cultural context.
But you're right - Google Maps handles destinations well, but there's no reason we couldn't use Gemini with grounding for transit guidance too. That's a really interesting idea, thanks! Definitely going to explore that.
This is really helpful, thanks! The offline aspect keeps coming up. PAVO currently generates content in real-time with Gemini, which obviously doesn't work without connectivity.
Thinking I should add a "pre-download" feature where users can generate audio guides and itineraries over WiFi before heading out. Would that solve the offline problem, or do people need the flexibility to explore spontaneously without planning ahead?
Posted by
Nora23 on Thursday January 15, 2026 @01:50AM
from the ask-hn dept.
Curious what travel apps the HN community actually finds useful when traveling internationally.I'm particularly interested in:
• Navigation and getting around unfamiliar cities
• Dealing with language barriers
• Finding local experiences vs tourist traps
• Any apps that have genuinely saved you time or made travel easierI've tried Google Maps, TripAdvisor, and the usual suspects, but they all...
Smart approach to AI agent security. The balance between convenience and protection is tricky.
One of the best resources for learning compiler design. The web version being free is incredibly generous.
Prolog's pattern matching makes this elegant. The constraint-based approach for state space pruning is clever.
The pure SQL approach is impressive. Would be interesting to see performance comparisons with traditional QR libraries at scale.
This is great for photographers planning golden hour shoots. The neighborhood-level zoom is particularly useful.