32M, salary 85k, Mumbai. Every month between 25th and 30th i’m stretching thin. Not overspending exactly, just timing - big grocery run, medical bills for parents, kid’s school stuff all seem to hit at the wrong week. Not broke broke. Salary lands on 1st and everything’s fine again. Just the 5-day squeeze is stressful. How do you all handle this? Personal loan feels overkill for 5 days.
Be honest. Most "agent eval" I see in the wild (including ours until recently) is a spreadsheet of prompts someone runs manually before big changes. That's not a regression suite. That's a vibe check with extra steps.
A real regression suite, the way we have for normal software, would mean: versioned test cases, runs automatically on every change, fails the build on regression, tracks pass-rate over time, and grows when new failure modes are found.
I want to know who's actually doing this for agents, and what it took to get there. Because the gap between "spreadsheet of prompts" and "real regression suite" feels large and I'm trying to figure out if it's worth crossing or if everyone's secretly still on spreadsheets.
i’m tired of hair advice that is basically “buy my code”.
so this is my current realistic Dubai hair stack, built for oily scalp + dry ends + hair fall panic:
blu shower filter - not expecting miracles, just hoping less rough/sticky hair
gentle shampoo - scalp only
conditioner/mask - only lengths, never scalp
Be Bodywise hair growth serum - night scalp step
Tangle Teezer - detangling without violence
Decathlon microfibre towel - because normal towels are rude
Bioderma SPF - on parting/hairline if I’m outside long
Nizoral - only if dandruff is active, not random daily use
this is probably the least sexy routine ever but maybe boring is the answer.
what would you remove/add?
32M, salary 95k. Have Axis Vistara + HDFC Millennia.
Want to add a third card just for offline UPI (groceries etc).
thinkin:
Kiwi
Tata Neu Plus
SuperMoney
Any real user experience with these?
I’m less worried about connecting Twilio Media Streams now.
That part is annoying but understandable.
What I’m stuck on is transcript state.
Live agent pipeline:
Twilio Media Streams
→ backend
→ STT
→ LLM
→ tool/CRM
→ TTS
→ audio back to caller
The dangerous bit is what happens between partial and final transcripts.
Example:
partial says: “book it for four”
LLM starts preparing 4:00
final says: “book it for four thirty”
oops
Or:
caller says: “my number is 9811… wait, sorry, 8911…”
partial catches the first one
final fixes it
CRM already saved the wrong value
So my current plan is:
- partials can update UI / detect rough intent
- finals can trigger actions
- numbers/dates require confirmation
- Redis stream per call
- cancel TTS on barge-in
- never let duplicate partials trigger duplicate tool calls
I want to try a Twilio Media Streams → Smallest AI Pulse → LLM flow because Pulse is positioned as real-time STT, but the make-or-break part for me is not just transcript accuracy.
It’s whether the event stream is predictable enough to build state around.
How are people handling partial/final transcript logic in Twilio voice agents?
was on tradetron about 8 months for my options strategies. fine to start but i kept hitting the ceiling, couldnt customize error handling, no visibility into actual execution, felt like i was fighting the platform.
want to go full api with python now. which broker is actually good to build your own stack on? care more about clean api and stable execution than brokerage.
need fridge and washing machine for a rented flat. staying in noida for now but office situation is not fixed.
buying used seems cheaper but if machine stops working then repairs are on me. renting feels easier but not sure about service and pickup later.
anyone here did this for short stay?
need fridge and washing machine for a rented flat. staying in noida for now but office situation is not fixed.
buying used seems cheaper but if machine stops working then repairs are on me. renting feels easier but not sure about service and pickup later.
anyone here did this for short stay?