Krishna Chaitanya Balusu
Kc · Member of Technical Staff at AIUC · San Francisco
About
I work on whether AI agents, and the evaluations that grade them, can be trusted. I build graders, retests and evidence records, and I look for the places where they fail silently.
I am a Member of Technical Staff at AIUC, on autonomous evaluations. Before that I spent four years at Meta on agent evaluation, Instagram Feed latency and data center chain of custody, and four years at Microsoft. In open source I work mostly on OpenTelemetry, where I wrote the generative AI plan operation.
NowAIUC, Member of Technical StaffAutonomous evaluations: checking that the tests used to certify AI agents grade them correctly. Since November 2026.
Publications
2026
Pinned Release Replays Expose Diagnostic Changes in AI Instrumentation
2026
Direct Pytest Assertions Localize GenAI Span Convention Drift
ISSRE 2026 Workshops (ReSAISE)code
2019
PowerApps Accessibility Standards and Guidelines
Microsoft white paperwith A. J. Gaud and S. Manasvi
Open source
OpenTelemetry GenAI plan operationsemantic-conventions-genai #97
A standard span for an agent's planning step. Support has since been added in Rust genai, Grafana, Tencent bk-monitor, Mistral and Mozilla.ai tooling.
Deprecating OpenCensus compatibilityOpenTelemetry Specification #5138
Released in spec v1.58.0, announced on the OpenTelemetry blog.
Collector memory limiter fixopentelemetry-collector #15053
Fixed a degenerate garbage collection loop first reported in 2022. Shipped in Collector v1.62.0.
Native Prometheus histograms in Istioistio #60350
Passthrough for native histograms, in the 1.32 prereleases.
OpAMP disconnect over plain HTTPopamp-spec #353
A spec change so agents on HTTP can disconnect cleanly.
METR Hawk evaluation fixesMETR/hawk #2654, #2809
Errored evaluation samples had been exported as successes.
Hugging Face Transformersv5.5.0 release notes
Config type annotations for strict validation; listed under significant community contributions.
Experience
AIUC Member of Technical Staff2026–
Autonomous evaluations: checking that the tests used to certify AI agents grade them correctly.
Meta Software Engineer2022–26
- Agent evaluation and observability2025–26Graders and human ground truth for OpsMate, Meta's incident investigation agent, and monitoring that shows when the grading itself breaks.
- Instagram Feed latency and reliability2024–25Wrote the Feed latency defense plan and per-launch regression thresholds, ran weekly regression reports, and shipped tail load and Reels in Feed reliability fixes.
- Data center chain of custody2022–23Reconciled custody records for data-bearing storage media with the hardware inventory, so each drive can be traced to its destruction.
Microsoft Software Engineer2018–22
Teams external access and PowerApps. One of three writers of the PowerApps accessibility guidelines.
Edelweiss Financial Services Machine learning intern2017–18
Talks & writing
Deterministic Outcomes in Non-Deterministic SystemsFeb 2026AI Agents Event East
Service
Mentor, SPAR Fall 2026A validity audit of trace-based failure monitors for LLM agents
Program committees and reviewingIEEE BigData 2026; workshops at NeurIPS, COLM and EMNLP
Member, OpenTelemetry organizationSince 2026
Fellow, BCS · Senior Member, IEEE
Education
Georgia Tech2023MS Computer Science, Computing Systems
BITS Pilani2018MSc Mathematics, BE Mechanical Engineering