WORK

What I’ve built.

The projects, the systems behind them, and the results.

Download résumé PDF

Onehouse

Senior Software Engineer

February–October 2023

Spark Agent

An AI assistant for debugging live Spark workloads.

Built end to end

I was the sole engineer building Spark Agent: the Scala/Java driver plugin, a supervised Rust subprocess, and the React interface. The two-process architecture communicated over versioned Unix-domain-socket IPC, with multi-provider LLM orchestration and 34 tools spanning the driver and Spark History Server. The assistant was publicly launched as part of Quanton.

Designed for the running system

A debugging assistant has to coexist with the workload it investigates. I validated runtime impact with TPC-DS and listener overhead with JMH microbenchmarks: measured wall-clock impact was −0.11%, with a p99 listener cost of 233 μs.

Protecting the knowledge layer

I encrypted an 82-document knowledge base using AES-256-GCM, added per-file canary sigils and CI-gated leakage tests, and implemented citation-path validation and per-tenant prompt watermarks.

Explore Spark Agent

Incident-response agent

Give on-call engineers a first investigation when production alerts fire.

From alert to investigation

I led a new AI team to build and ship the root-cause-analysis agent, owning architecture, LangGraph orchestration, MCP tool integration, and access control. We used a runbook-gated deterministic flow with a bounded non-deterministic fallback.

At the time of this work, the agent handled first-level investigation for all production Sev1 alerts, with 60–70% accuracy. I also built secure dataplane connectivity across AWS and GCP.

Making investigations economical

I combined response trimming, tool-history compaction, and prompt caching to reduce the token cost of an investigation by approximately 30%.

Platform infrastructure

Took the infrastructure from its initial foundation to multi-cloud operations.

Automating the foundation

I built the infrastructure foundation as the sole engineer and migrated customer environments to Terraform automation. CI/CD brought deployment time from two days to one hour, while self-hosted GitHub runners reduced operations costs by 40%.

I established a shared development environment using Terraform Cloud and GitHub, with resource tracking, TTL-based cleanup, and auditability. This reduced cloud costs by 30%.

Orchestration and multi-cloud

I designed Java services for Terraform module updates and customer onboarding, including a Redis-based workflow engine for Kubernetes cluster creation. The platform supported private clusters, Calico networking, and custom AMIs.

I led the GCP control-plane rollout from an AWS-only platform: secure infrastructure patterns, API hardening, external penetration testing, disaster recovery, and achieving SOC 2 compliance.

Scaling against actual demand

I led the RFC, architecture, and rollout for Spark-workload-aware EBS volume autoscaling, including evaluating NVMe alternatives. The initiative reduced combined EC2 and EBS costs by 45% for a large customer.

I also designed a Java-based Spark router and an autoscaling algorithm that scaled Spark applications according to query and resource demand.

Building the operating rhythm

I led infrastructure and on-call standups, unblocked engineers across teams, and drove improvements that reduced weekly Sev1–Sev3 alert volume from approximately 250 to 60. Through reviews, incident troubleshooting, and reliability practices, I mentored engineers as the infrastructure team grew from one to six.

Explore Onehouse Cloud

GitHub

Software Engineer II

September 2021–October 2022

npm security and data exports

Make npm accounts safer and give users controlled access to their data.

Safer experiences on npm

I built a JavaScript SDK for Octocaptcha and integrated the captcha framework into multiple npmjs.com user-facing flows. I also built a GDPR-compliant data export pipeline using AWS SQS, Lambda, and EC2 with scoped IAM permissions.

My security work included JWT secret rotation without traffic impact and MFA enforcement for package maintainers. I worked through accessibility audit findings across npm surfaces.

Azure build infrastructure

Get development pipelines running faster.

Faster pipelines and better visibility

I compared multiple agent configurations before replacing default Azure pipeline agents with custom VMSS-based agents, improving pipeline performance by 50%. I also built monitoring for new instance creation using Geneva metrics and ICM.

Adobe

Intern → Senior Member of Technical Staff

Intern: May–July 2017 · Full-time: June 2018–January 2021

Acrobat Liquid Mode

Make PDFs easier to read with adaptive layouts.

I developed an adaptive PDF reading experience using content classification and extraction. I tuned ML model parameters using a genetic algorithm in C++, improving precision for good and bad pages by 1% and 5%, respectively.

I automated model-checkpoint selection in Python, reducing manual effort by 66%, and owned component deployment pipelines covering code review, model registration, and end-to-end verification.

Read about Liquid Mode

Adobe Connect

Bring collaboration features to accessible web and mobile experiences.

I led mobile app development, implemented SSO, upgraded the Adobe AIR SDK, and shipped iOS version 2.7.0. I also migrated file-sharing and web-link features from Flash to HTML with React and Java, including full accessibility support. I built a unified recording interface with playback controls using React Spectrum.

Click prediction

Extend bidding coverage to keywords with limited historical data.

Adobe’s bidding system needed to predict costs and returns for keywords with little historical data. I proposed clustering these sparse keywords using features from data-rich keywords, then built a model to estimate their probability of receiving clicks. This extended the bidding system’s keyword coverage.

Flipkart

Software Development Intern

May–July 2016

Understanding customer text

Find useful structure in customer queries and emails.

I proposed a domain-specific chunking grammar to identify useful phrases in customer queries and inbound and outbound emails. I combined it with ReVerb, Ollie, and Stanford CoreNLP for information extraction, then used word2vec embeddings and Hierarchical Agglomerative Clustering to group the extracted phrases.