~/talk

Navigating Real-World Challenges in a Production-Grade Multi-Agent System

linkedin:sibinbh
slides:none on file

A production conversational assistant for cricket data often fails on specifics: an observed Agent Kohli failure answers ODI World Cups with IPL data. The build uses a text-to-SQL pipeline validated via text-to-SQL deterministic checks (inspired by Defog SQL Eval), Pydantic-structured outputs, Postgres RBAC and API JWT guardrails, and tool trials like Duckling (insufficient). Prompt compression reduces token size ~60% prompt compression, runtime monitoring logs every LLM call and a runtime quality flag marks answered/partial/unanswered interactions for debugging.

AI-written summary · about

  1. 0:00Introduction to multi-agent systems
  2. 3:12High-level system architecture
  3. 4:17Core aspects for production readiness
  4. 5:41Testability in agentic apps
  5. 8:54Accuracy improvement strategies
  6. 15:04Prompt engineering best practices
  7. 19:30Reducing end-to-end latency
  8. 22:37Cost optimization techniques
  9. 24:50Observability and debugging
  10. 27:42Key takeaways and conclusion
~/search
type to search 6,518 conferences, 143,918 speakers and 192,133 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
← →
on a talk: 5 seconds back / forward
0–9
on a talk: jump to 0%–90% of the recording
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.