Software engineer working on AI infrastructure, LLM inference observability, Kubernetes, and backend systems in Go and Python. Based in India.

• Kubernetes-native diagnostics for LLM inference (vLLM, Triton)
• Observability with Prometheus and OpenTelemetry
• Backend services: FastAPI, PostgreSQL/pgvector, REST APIs

p95labs/ifa · Creator & Maintainer · Aug 2026 – present · independent open-source project, alpha

A read-only Go diagnostics layer for LLM inference on Kubernetes. It correlates Prometheus runtime telemetry with cluster state (via client-go) and produces 19 coded, evidence-backed findings across 7 rule families.

• Validated against live vLLM 0.28.0 (CPU and one NVIDIA L4), Triton 25.12 (CPU only), and a DCGM-compatible exporter on the L4. Kubernetes discovery is tested end to end in a kind cluster (Kubernetes 1.31) in CI
• Read-only /api/v1/* API and ifa CLI. Docker + Helm packaging with get/list/watch-only RBAC
• Missing telemetry is reported as missing, never treated as zero
• Removed components that couldn't be defended (a no-op node agent, an invalid Ollama path, static sample outputs) and documented the remaining validation gaps

Personal Knowledge Vault: a deployment-oriented FastAPI + PostgreSQL/pgvector RAG service with JWT auth, cited answers, Alembic migrations, Docker Compose, GitHub Actions CI, and a retrieval evaluation harness (Precision@K, Recall@K, MRR).

• LlamaIndex #20767 (merged): fully local RAG notebook with Chroma and Ollama
• LangChainGo #1506 (open): OpenTelemetry callback handler in Go
• OpenLLMetry #4227 (open): FastAPI + LiteLLM tracing sample app
• Haystack Cookbook #272 (open): advanced local RAG notebook
• Ragas #2605 (open): AnswerConsistency metric
• Future AGI #1964 (open): surface all key-validated providers in the Add Provider dialog

Go · Python · SQL · Bash · Kubernetes · client-go · Helm · Docker · kind · GitHub Actions · Prometheus · OpenTelemetry · vLLM · Triton · FastAPI · PostgreSQL · pgvector

p95labs.com · LinkedIn · prajwalmoras10@proton.me