LFX Mentee · CNCF / Jaeger
I build, evaluate and trace AI agents.
LFX Mentee at the CNCF, working on Jaeger. I built Jaeger's MCP skills framework, made Jaeger trace its own agents, and found out why agents skip the tools you give them.
trace a3f1c9e2d47b8e05 · 28 spans · 7 services · 1.80 s
scroll to open the laptop
scroll ↓
46
merged PRs
in other people's repos, as of 10 Oct 2026
28
merged in Jaeger + OpenTelemetry
jaeger, jaeger-ui, otel-collector, otel-collector-contrib
0/5 → 5/5
skill discovery
two agent harnesses, after rewriting tool descriptions
3
agent sidecars
Gemini (Google ADK), Claude, Goose on one framework
Featured work · Jaeger
LFX Mentorship, Jun 2026 – Sep 2026. Mentors: Yuri Shkuro (Jaeger creator) and Jonah Kowall. Built the Jaeger AI Self-Service Skills Framework: an MCP tool that lets AI agents load expert trace-diagnosis playbooks on demand, an agent-agnostic gateway over ACP, and the tracing that lets Jaeger observe its own agents.
Three-level progressive disclosure: catalog → sub-skill → references. An agent loads only the playbook it needs.
Fixed WebSocket message framing so any ACP agent connects, then added authenticated agents via configurable handshake headers.
OpenTelemetry GenAI span attributes plus W3C traceparent across ACP. Gateway and sidecar spans join into one trace per chat turn.
Agents skipped skills because they judged direct telemetry tools sufficient: a reasoning problem, not a discovery problem.
Labeled trace fixtures and end-to-end coverage for the built-in skills. Operator skills served from ai.skills_dir.
Two jaeger-ui PRs: expandable tool-call blocks in the chat thread, Markdown rendering and panel resize.
Open source
46 merged pull requests in 6 repositories, 29 of them in CNCF projects. Counted on Oct 10, 2026.
CNCF · graduated · Go
MCP skills framework, ACP gateway, GenAI tracing, SigV4 and custom-header auth for Elasticsearch/OpenSearch and Prometheus backends, OTLP fields in memory storage, schemagen for internal extensions.
CNCF · graduated · TypeScript / React
Trace Logs view aggregating span events; tool-call rendering, Markdown and panel resize in the assistant.
CNCF · graduated · Go
Prometheus remote-write exporter: disable_scope_info and scope-label translation; instrumentation scope attributes as labels; Elasticsearch receiver IPv6 endpoints; Elasticsearch exporter require_data_stream.
CNCF · graduated · Go
mdatagen: preserve custom extensions during schema reference resolution.
CNCF · graduated · Go
Build cleanup: removed unused registry build paths.
open source · Go
Memory engine, trust and policy layers, API integration, governance UI scaffold, ADRs, CI and DCO enforcement.
Most recent merges
Project
Agentic RAG over any public GitHub repo
Ask plain-English questions about an unfamiliar codebase. A LangGraph agent clones and ingests the repo, chunks it with language- and symbol-aware splitting, and retrieves with ChromaDB and sentence-transformer embeddings.
Stack
repo → clone → chunk (language + symbol aware) → embed → ChromaDB → LangGraph agent → answer / Mermaid
Skills
Education
B.Tech, Information Technology, Jabalpur Engineering College · expected Apr 2027
Certifications
About
The same view Jaeger gives a span: tags, then the log of what happened.
I'm a final-year IT student at Jabalpur Engineering College and an LFX Mentee at the CNCF, working on Jaeger under Yuri Shkuro (Jaeger's creator) and Jonah Kowall. I built the Jaeger AI Self-Service Skills Framework: an MCP tool that lets any AI agent load expert trace-diagnosis playbooks on demand, and the plumbing that makes Jaeger trace its own agents.
I got into observability by troubleshooting a slow service while learning DevOps, and stayed because the question "why did this agent do that?" turns out to be the same question as "why is this request slow?". Outside code: swimming and writing poems.
TODO: photo
slot reserved · the monogram is the only mark until then
Tags
Logs
Contact
Open to roles in agentic AI, evals and observability from 2027. Open-source collaboration any time.
