status: 200 OK ·San Jose, CA ·lat 37.34 · lon -121.89

Jayanth
Yarlagadda

<software_engineer />

I specialize in backend infrastructure — ingestion pipelines, distributed systems, and the observability layers that keep them reliable in production. Recent work includes a multi-tenant security telemetry pipeline at Microsoft, alongside open-source contributions: GPU-accelerated tooling for NVIDIA RAPIDS and infrastructure improvements across AWS and OpenAI.

View Experience Download Résumé GitHub

Experience

Software Engineering Intern

Microsoft — Security Platform & AI (SP&AI)

Redmond, WA May 2026 – Aug 2026
  • Productionized FluxBridge, a multi-tenant security data-lake ingestion pipeline on AKS, enabling Vector workers to process protobuf-encoded security telemetry into Parquet across 50–80 input streams.
  • Built an end-to-end OpenTelemetry plane reshaping internal metrics and logs into OTLP, tagging every signal with worker, schema, and service metadata for per-stream debugging.
  • Deployed an in-cluster OpenTelemetry Collector gateway forwarding telemetry to App Insights, Log Analytics, and Azure Grafana — validating 98K+ metric rows/hour across 8 dashboards.
  • Hardened the AKS deployment for compliance-constrained environments, packaging a C# ingestion worker and a Rust-patched Vector binary into an ACR-hosted image with per-stream autoscaling.
  • Engineered a C# drain-detection mechanism correlating Prometheus metrics to eliminate indefinite worker hangs; built a Go remediation component driven by throughput and error-growth thresholds.
  • C#
  • Go
  • Rust
  • Kubernetes (AKS)
  • OpenTelemetry
  • Azure Monitor
  • Prometheus
  • Grafana

Cloud Intern

AWS Academy Cloud Virtual Internship

Remote May 2023 – Jul 2023
  • Architected a secure 2-tier AWS deployment with EC2, S3, VPC, and Route 53 — implementing subnet isolation, route-table segmentation, and security-group controls.
  • Hardened IAM access and load-tested EC2 configurations, reducing over-permissioned access by ~40%, response time by ~20%, and projected cost by ~15%.
  • AWS
  • EC2
  • S3
  • IAM
  • VPC
  • Route 53

Open Source Contributions

microsoft / PyRIT PR #2884

Speed up GCG candidate evaluation

Profiled the GCG candidate-evaluation path and eliminated full-vocabulary logits per candidate — computing cross-entropy inside each model worker, projecting only needed positions via logits_to_keep, and optionally reusing the prefix KV cache across the batch.

~31% faster median step Python LLM red-teaming
View pull request
aws / aws-lambda-runtime-interface-emulator PR #185

Fix RIE init duration reporting

RIE stopped measuring init time as soon as the init handler returned. Diagnosed and fixed the timing boundary so cold-start metrics reflect real init duration.

Go AWS Lambda Cold-start timing
View pull request
rapidsai / cudf PR #23286

cudfgrep — GPU-accelerated grep

Authored a Unix grep-style CLI on top of cuDF's GPU string engine for massively parallel regex search — case-insensitivity, inverted matches, multi-pattern search, and correct grep exit codes.

~20 GB/s GPU scan (T4) ~8–9× faster than ripgrep CUDA / cuDF string kernel
View pull request
openai / harmony PR #148

Fix quadratic BPE merging for long pieces

The tokenizer's byte-pair merge degraded to O(mn) on long pieces. Replaced it with a binary heap over linked boundary arrays for large inputs, cutting merge work to O(m log n) while keeping the fast path for short pieces.

O(mn) → O(m log n) Rust
View pull request
NVIDIA / garak PR #1955

Persuasive Adversarial Prompt probe

Implemented probes.persuasion.PersuasiveJailbreak for NVIDIA's LLM vulnerability scanner, based on the "How Johnny Can Persuade LLMs to Jailbreak Them" research (Zeng et al., arXiv:2401.06373).

LLM red-teaming Python
View pull request
openai / tiktoken & harmony PRs

Tokenizer error handling & API ergonomics

Added a dedicated DisallowedSpecialTokenError to tiktoken so callers can catch disallowed special tokens without parsing strings, plus re-exported RenderConversationConfig from harmony's crate root for Rust callers.

Python Rust
View pull request

Independent engineering work

Orbit — durable edge command delivery

A durable command-delivery system for intermittently connected edge devices, built around a verification loop: a deterministic C++20 fault engine reproduces failure scenarios against the real Go services, and a history checker verifies the documented guarantees.

At-least-once + fencing tokens Deterministic fault replay
  • Go
  • C++
  • Distributed Systems
View repository

Vyomel — AI execution platform

A personal AI execution platform that turns natural-language intent into verified, permission-aware actions across browser, desktop, and API tools — with a focus on guardrails and verification rather than blind automation.

Permission-aware actions Browser · desktop · API tools
  • Python
  • LLM Agents
  • Automation
View repository

Computer-Use Automation

An LLM discovers a legacy back-office UI flow once, then compiles it into a typed, versioned capability artifact that replays deterministically — with no model left in the decision loop at runtime.

Record-once, replay deterministically No model in the runtime loop
  • TypeScript
  • LLM Agents
  • Automation
View repository

Production-Ready Cloud Infrastructure

A production-ready cloud deployment pipeline pairing a FastAPI service with Terraform-provisioned AWS infrastructure, containerized with Docker and shipped through a GitHub Actions CI/CD workflow.

Terraform-provisioned AWS Dockerized CI/CD
  • Python
  • Terraform
  • Docker
  • AWS
  • GitHub Actions
View repository

Mobile Edge Computing — Task Offloading

Led a 3-person team designing an iFogSim simulation framework evaluating task offloading between edge devices and cloud nodes, with an interactive UI for scheduling and monitoring.

+25% resource allocation 3rd Prize, Project Expo 2024
  • Python
  • iFogSim
  • AWS
View repository

Secure Healthcare Data Access & Search Framework

Co-engineered a healthcare records portal using searchable encryption, enabling secure querying over encrypted patient data without ever exposing sensitive information. Led front-end design and collaborated on the secure backend architecture on AWS.

Searchable encryption ~18% lower query latency
  • Java
  • AWS
  • MySQL
View repository

Toolbox

Languages

  • Python
  • Java
  • C++
  • C#
  • Go
  • Rust
  • JavaScript
  • TypeScript
  • SQL
  • HTML/CSS
  • Bash

Cloud & Infrastructure

  • Azure
  • AKS
  • Kubernetes
  • AWS
  • EC2
  • S3
  • IAM
  • VPC
  • Terraform
  • Docker
  • Jenkins
  • GitHub Actions
  • CI/CD

Backend & Observability

  • OpenTelemetry
  • OTLP
  • Vector
  • Prometheus
  • Grafana
  • App Insights
  • REST APIs
  • Spring Boot
  • Flask
  • Node.js
  • Express.js

Data & Systems

  • NVIDIA cuDF
  • CUDA
  • Protobuf
  • Parquet
  • MySQL
  • PostgreSQL
  • MongoDB
  • Redis
  • Kafka
  • pytest
  • xUnit

Academic background

San José State University

Master of Science in Software Engineering

Jan 2025 – Dec 2026 · San Jose, CA · GPA 3.6/4.0

SRM Institute of Science and Technology

B.Tech, Computer Science & Engineering (Cloud Computing)

Aug 2020 – May 2024 · Chennai, India · CGPA 8.5/10

Or just ask the terminal

It actually works — type help to see what it can do.

jayanth@portfolio:~
Welcome. Type help to see available commands.
jayanth@portfolio:~$

Let's build something reliable together

Open to Software Engineer opportunities. The fastest way to reach me is email.

Email jayanthyarlagadda@gmail.com LinkedIn in/jay-yarlagadda GitHub @JayYarlagadda