Practical observability with Python
Connect logs, metrics and traces through runnable examples.
Read the series ↗Soumendra Kumar Sahoo
AI Observability Architect · Hyderabad
Building AI
people can
trust.
I help teams understand AI behavior and make sound operating decisions. I build with AI agents, setting clear boundaries so the people who inherit the work can understand and maintain it.
Sharing what I learn.
AI Day · Bhubaneswar · personal archive
Understand the system.
Make the next decision.
My work connects what an AI system does with what its operators need to know: what to instrument, how to evaluate behavior and how to respond when things go wrong.
At PepsiCo, I lead AI observability architecture. Earlier work on ML platforms at Freshworks and AIOps at IBM shaped how I approach reliability, access and the teams maintaining a system.
Read the engineering story ↗One way I put that into practice
Make AI behavior
easier to inspect.
Widescope
A browser-native viewer I build for LLM and AI-agent traces. Open a captured run and follow its timing, span relationships and AI calls.
I chose local inspection: your trace stays in your browser, with no ingestion service to operate.
Understanding should
travel beyond one person.
I turn implementation experience into practical writing, runnable examples and sessions that other engineers can use.
Explore my AI observability work ↗What does evaluation cost?
Consider what you inspect, what you spend and what a sample might miss.
Read the article ↗In the room: AI Observability SDK session · PyConf Hyderabad, 2026. Talks and session resources ↗
“…exceptional documentation skills and attention to detail, which made it easy for the team to understand and build on his work effectively.”
Krithika Arunachalam · Senior Data Scientist, Freshworks
Read the full colleague recommendation ↗