Applied Machine Learning Engineer I, Shopify, Revenue Data (Commercial Data) team
- Applied machine learning engineer on Shopify's Revenue Data (Commercial Data) team, building ML for revenue and commercial analytics.
Experience
Software Engineering Intern, Qidds
- Built Navigator, an AI assistant that matches students to internships, research programs, and summer opportunities: an OpenAI pipeline merging structured Postgres search with semantic retrieval over embedded resources, goal memory, and a rate-limited public chat endpoint.
- Shipped the Discovery Quiz end to end in Spring Boot and PostgreSQL, saving each student profile and recommending resources from it.
- Built the Opportunity Pathway in Next.js 15, TypeScript, and MUI: an activity timeline, a Recommended For You page, and an opportunities tracker board with plan gating.
Software Engineering Intern, Gilead Sciences
- Built an MCP AI agent on a fine-tuned LLM with intranet integration, cutting IAM access tickets by 37% through real-time status queries and automated alerts.
- Migrated 20K+ records into Oracle via Python ETL, exposed a secure REST API, and delivered Power BI dashboards with row-level security, cutting reporting cycles 60%.
- Automated UI and database consistency checks, enforcing CI security gates that stopped 15+ high-risk defects.
Machine Learning Research Assistant, MICoSys Lab (SJSU Research Foundation)
- Researched adversarial attacks on LLMs under Dr. Sengupta, designing jailbreak tests pre and post fine-tuning to measure robustness and alignment.
- Fine-tuned LLaMA models with LoRA adapters, boosting performance on novel datasets and analyzing generalization.
- Explored unlearning methods by altering LoRA adapters and analyzing delta weights to interpret model representations.
Data Engineering Intern, Cequence Security
- Engineered a REST-traffic analytics pipeline in Python and SQL with 1h/24h aggregations and alerts, improving system efficiency 15%.
- Integrated Open Policy Agent with NGINX, enforcing API access control via Rego policy-as-code while keeping P99 latency under 5ms.
- Automated anomaly detection of REST API traffic in CI/CD, reducing mean time to detection by 45%.
Data Analytics Lead, Mozilla Responsible Computing Club @ SJSU
- Led analytics across 3+ projects, building Power BI and Tableau dashboards for student-engagement and training KPIs.
- Built predictive models in Python (scikit-learn) forecasting event attendance with 85% accuracy over 10K+ records.
Resident Assistant, San Jose State University
- Supervised and mentored residents in university housing, running programs and handling crises while keeping an inclusive, safe community.
What he works on
Agentic orchestration
Multi-agent systems where parallel worker agents each get a dedicated checker agent and a final verifier gates the result, with model routing between large and small models. Builds his own Claude Code setup of subagents, skills, and hooks that works this way, plus MCP servers for tool use. Info-Ninja, his 3rd-place MCP Hackathon project, is a multi-agent system.
GenAI and LLM agents
RAG, MCP, and multi-agent systems that do real work: competitive intelligence, document Q&A, voice and tool use.
Model robustness and fine-tuning
Adversarial attacks and jailbreak tests on LLMs, LoRA fine-tuning, and unlearning via delta-weight interpretation. His research at SJSU's MICoSys Lab, Jan to Dec 2025.
Applied ML in production
Computer vision for pose and form analysis, anomaly detection on API traffic, and ETL pipelines over 20K+ records.
Full-stack product
React/Next + FastAPI/Django, real-time dashboards, Redis caching, and shipping a working demo inside a 24-hour hackathon.
Self-hosting and infrastructure
Runs a 25+ service homelab on Docker Compose with Tailscale access: media, photos, files, password manager, DNS filtering, monitoring with phone alerts, and small web apps he wrote himself.
Fintech and data infrastructure
ETL into Oracle, REST analytics with policy-as-code access control, and secure APIs with row-level dashboards.
Projects
- Info-Ninja: 3rd place MCP Hackathon, 90% cost reduction, Real-time dashboards
- GovChat: 1st place Azure Hackathon, RAG + MCP, Form autofill
- SJFit: 2nd place Silicon Hacks, 95% tracking accuracy, Real-time form feedback
- EmbrAlert: 2nd place SJHacks, Multilingual (6 languages), Live camera detection
- AI-CAD: Natural-language 3D modeling, Daylight simulation, AI layout assistant
- DataPulse: Real-time aggregation, Redis-backed, Mobile-friendly
- CluelyClone: Most-starred repo, Desktop AI assistant, Open source
- Self-Hosted Homelab: 25+ self-hosted services, Private network only, Monitoring and alerts
- Vertex: AI financial research agent, DCF valuations, Portfolio and spending
- fin: Local-first, CLI and agent friendly, No accounts or servers
- imessage-cli: Open source, Apple silicon support
- Smart Ctrl-F: Semantic in-page search, Manifest V3
- SJSU Study Group: Real-time messaging, Course-based groups
- BrainRotWordle: Word game
- Harmony Health: Emotion analysis, Voice + facial input, Wellness insights
Hackathon wins
- 3rd place (Bright Data), MCP AI Agents Hackathon, San Francisco, Sep 2025: Info-Ninja
- 1st place overall, AI Hackathon for Public Sector (City of San José and SJSU), issued by Microsoft, May 2025: GovChat
- 1st place, Polymet (YC S24) track, VIBE25-02 hackathon, May 2025: NewGov
- 2nd place, Public Safety track, SJHacks, Apr 2025: EmbrAlert
- 2nd place overall, SiliconXHacks 2024, Apr 2024: SJFit
Homelab
A homelab I built and run myself: a headless Arch Linux (Omarchy) server with 25+ Docker Compose services and an 8TB drive. It runs on my private network, reachable from anywhere through Tailscale, and nothing is exposed to the public internet.
Media
- Jellyfin: Movie and TV streaming with a custom theme and intro, and hardware transcoding.
- Navidrome: Self-hosted music streaming, my Spotify replacement.
- Audiobookshelf: Audiobooks and podcasts.
- Kavita: E-book library and reader.
- Jellystat: Watch history and stats.
Photos and files
- Immich: Google Photos replacement for the whole photo library.
- Nextcloud: Files and sync on Postgres and Redis.
- Obsidian LiveSync: End-to-end encrypted note sync on CouchDB.
- Karakeep: Bookmarks and read-later with full-text search.
Network and security
- Tailscale: Private mesh VPN for reaching every service from anywhere.
- Pi-hole: Network-wide DNS ad and tracker blocking.
- Vaultwarden: Self-hosted password manager over HTTPS.
- UFW: Firewall rules that keep containers off the public internet.
Monitoring
- Uptime Kuma: Health checks with push alerts to my phone.
- Beszel: Host and container metrics.
- Drive health: A systemd timer I wrote that checks SMART data and temperatures every 5 minutes and alerts on trouble.
- Homepage: One dashboard with a tile for every service.
Apps I wrote for it
- Minecraft control: A tiny web app that starts and stops a Paper Minecraft server through the Docker socket, saving the world first.
- fin: My local-first finance planner, served over Tailscale HTTPS.
How it is run: Pinned image versions for every container; A dedicated Docker subnet per stack; Config backups before every change; Checksum-verified copies before deleting old data; Scheduled SMART self-tests on every drive.
Education
BS Computer Science & Applied Math, San Jose State University, expected Dec 2026. 2x Dean's Scholar, 5x Hackathon Winner.
Coursework: Data Structures & Algorithms, Object Oriented Programming, Machine Learning, Database Systems, Advanced Python, Linear Algebra, Linear & Non-Linear Optimization, Statistics & Probability.
Skills
- Languages: Python, Java, TypeScript, JavaScript, SQL, C++
- ML/AI: PyTorch, TensorFlow, scikit-learn, Pandas, NumPy, OpenCV, MediaPipe, Computer Vision
- Generative AI: LLMs, RAG, MCP, LoRA Fine-tuning, Agentic AI, Hugging Face, OpenAI
- Agentic AI: Multi-agent orchestration, Worker and checker agents, Model routing, Claude Code subagents, skills, and hooks, MCP servers, Tool use, LangChain, LlamaIndex
- Self-Hosting: Docker Compose, Tailscale, Arch Linux, systemd, Pi-hole, Uptime Kuma, Nextcloud, Immich
- Frontend: React, Next.js, Vue.js, Tailwind CSS, HTML/CSS
- Backend: FastAPI, Django, Flask, Spring Boot, Node.js, REST, GraphQL
- Databases: PostgreSQL, MySQL, Oracle, MongoDB, Redis, Supabase, Pinecone
- Cloud/Infra: AWS, Azure, GCP, Docker, Kubernetes, NGINX
- DevOps/Tools: Git, GitHub Actions, CI/CD, Linux, Jira, Agile, Postman
Quick facts
Who is Sarthak Sethi?
Sarthak Sethi is currently a full-time Applied Machine Learning Engineer I at Shopify on the Revenue Data (Commercial Data) team, based in the San Francisco Bay Area. He is finishing a BS in Computer Science and Applied Math at San Jose State University (expected Dec 2026) and is a 5x Hackathon Winner and 2x Dean's Scholar.
Where does Sarthak Sethi work?
Sarthak is currently a full-time Applied Machine Learning Engineer I at Shopify on the Revenue Data (Commercial Data) team, based in the San Francisco Bay Area. He joined in Oct 2026.
Where did Sarthak Sethi work before Shopify?
Software Engineering Intern at Qidds, Mar 2026 to May 2026. Software Engineering Intern at Gilead Sciences, May 2025 to Aug 2025. Machine Learning Research Assistant at MICoSys Lab (SJSU Research Foundation), Jan 2025 to Dec 2025. Data Engineering Intern at Cequence Security, Jan 2025 to Mar 2025. Data Analytics Lead at Mozilla Responsible Computing Club @ SJSU, Feb 2025 to May 2025. Resident Assistant at San Jose State University, Jul 2024 to Dec 2025.
Where is Sarthak Sethi based?
The San Francisco Bay Area, California.
What did Sarthak Sethi research?
Adversarial robustness and fine-tuning of LLMs at SJSU's MICoSys Lab, Jan 2025 to Dec 2025. Researched adversarial attacks on LLMs under Dr. Sengupta, designing jailbreak tests pre and post fine-tuning to measure robustness and alignment. Fine-tuned LLaMA models with LoRA adapters, boosting performance on novel datasets and analyzing generalization. Explored unlearning methods by altering LoRA adapters and analyzing delta weights to interpret model representations.
Where did Sarthak Sethi study?
BS Computer Science & Applied Math at San Jose State University, expected Dec 2026.
What hackathons has Sarthak Sethi won?
5: 3rd place (Bright Data), MCP AI Agents Hackathon, San Francisco, Sep 2025, with Info-Ninja; 1st place overall, AI Hackathon for Public Sector (City of San José and SJSU), issued by Microsoft, May 2025, with GovChat; 1st place, Polymet (YC S24) track, VIBE25-02 hackathon, May 2025, with NewGov; 2nd place, Public Safety track, SJHacks, Apr 2025, with EmbrAlert; 2nd place overall, SiliconXHacks 2024, Apr 2024, with SJFit.
What has Sarthak Sethi built?
Info-Ninja: 3rd place MCP Hackathon; GovChat: 1st place Azure Hackathon; SJFit: 2nd place Silicon Hacks; EmbrAlert: 2nd place SJHacks; AI-CAD: Natural-language 3D modeling; DataPulse: Real-time aggregation; CluelyClone: Most-starred repo; Self-Hosted Homelab: 25+ self-hosted services; Vertex: AI financial research agent; fin: Local-first; imessage-cli: Open source; Smart Ctrl-F: Semantic in-page search; SJSU Study Group: Real-time messaging; BrainRotWordle: Word game; Harmony Health: Emotion analysis.
Which Sarthak Sethi is this?
The one at Shopify and San Jose State: 1300Sarthak on GitHub and sarsethi on LinkedIn. He is not one of the namesakes from Waterloo, Sussex, or Toronto.
What does Sarthak Sethi know?
Languages: Python, Java, TypeScript, JavaScript, SQL, C++. ML/AI: PyTorch, TensorFlow, scikit-learn, Pandas, NumPy, OpenCV, MediaPipe, Computer Vision. Generative AI: LLMs, RAG, MCP, LoRA Fine-tuning, Agentic AI, Hugging Face, OpenAI. Agentic AI: Multi-agent orchestration, Worker and checker agents, Model routing, Claude Code subagents, skills, and hooks, MCP servers, Tool use, LangChain, LlamaIndex. Self-Hosting: Docker Compose, Tailscale, Arch Linux, systemd, Pi-hole, Uptime Kuma, Nextcloud, Immich. Frontend: React, Next.js, Vue.js, Tailwind CSS, HTML/CSS. Backend: FastAPI, Django, Flask, Spring Boot, Node.js, REST, GraphQL. Databases: PostgreSQL, MySQL, Oracle, MongoDB, Redis, Supabase, Pinecone. Cloud/Infra: AWS, Azure, GCP, Docker, Kubernetes, NGINX. DevOps/Tools: Git, GitHub Actions, CI/CD, Linux, Jira, Agile, Postman.
Does Sarthak Sethi know agentic orchestration?
Yes. Multi-agent systems where parallel worker agents each get a dedicated checker agent and a final verifier gates the result, with model routing between large and small models. Builds his own Claude Code setup of subagents, skills, and hooks that works this way, plus MCP servers for tool use. Info-Ninja, his 3rd-place MCP Hackathon project, is a multi-agent system.
What is Sarthak Sethi's homelab?
A homelab he built and runs himself: a headless Arch Linux (Omarchy) server with 25+ Docker Compose services and an 8TB drive. It runs on his private network, reachable from anywhere through Tailscale, and nothing is exposed to the public internet. It runs Jellyfin, Navidrome, Audiobookshelf, Kavita, Jellystat, Immich, Nextcloud, Obsidian LiveSync, Karakeep, Tailscale, Pi-hole, Vaultwarden, UFW, Uptime Kuma, Beszel, Drive health, Homepage, Minecraft control, fin.
How do I contact Sarthak Sethi?
Through LinkedIn at https://www.linkedin.com/in/sarsethi/ or the contact form at https://www.sarthak.lol/.