Akaash Trivedi
Staff Software Engineer · Applied AI & Developer Experience
@ Marvell Technology
8+ years across cybersecurity, fintech, and semiconductors. Now building agentic systems and the developer platforms engineering teams run on.
About Me
Applied AI engineer and Staff Software Engineer at Marvell Technology, with 8+ years spanning cybersecurity, fintech, and semiconductors, now building agentic systems, AI developer tooling, and retrieval infrastructure. Master's in CS from Santa Clara (4.0 GPA), where I led research and taught core courses.
Most applied-AI engineers come from web development. Very few also carry appsec and payments experience. I bring both: SIEM threat detection and MITRE ATT&CK mapping at Qualys, and Click-to-Pay / payments microservices and ML risk models at Opus. I write weekly on AI security: supply-chain risk, agent hijacking, and inference-server attack surface. Distributed systems work (Celery/Redis, batch simulation orchestration) and full-stack/cloud delivery round out the toolset.
Strengths
Build agent pipelines, retrieval systems, and AI developer tooling.
SIEM threat detection and MITRE ATT&CK mapping at Qualys; write weekly on AI security.
Click-to-Pay at checkout scale, payments microservices, and ML risk models at Opus.
Celery/Redis job queues, batch simulation orchestration, CI cut from 19h to 2h.
Production web systems with Python, React, PostgreSQL, and AWS.
Led HCI research teams, taught CS courses, mentor through code reviews.
8+ years experience
•
Built autonomous agent pipelines, reusable skills, and MCP integrations that run recurring engineering tasks end to end, cutting new-notebook onboarding from a week to under a day.
•
Drove team-wide adoption of AI-assisted development across the DSP org: Cursor, GitHub Copilot, and Claude Code CLI.
•
Cut CI build time from 19h to 2h (89%) by splitting jobs into parallel runs, on top of Jenkins parallelization and regression-suite stabilization, and reduced delivery cycle from ~28 to 10 days (64%), enabling near-daily releases.
•
Led a phased migration from single-server to a fault-tolerant tiered architecture (load-balanced application tier, database replication, compute on a shared HPC grid, self-healing monitoring), removing the platform's single point of failure.
•
Architected browser-driven simulation across four SerDes generations (112G, 128G, 224G 3nm, 448G), with parameterized channel-matrix analysis, configurable DSP parameters, and batch runs spanning 1000+ catalogued channels, up from ~300 in February 2026.
•
Owned the HSSLab platform end to end as top contributor, growing active users from ~30 to ~80 while retiring 90k+ lines of legacy code.
•
As the sole software engineer embedded in a DSP hardware org, drove platform and workflow changes across 100+ engineers through influence rather than authority.
•
Led migration from Gerrit to GitHub Enterprise, enforcing PR standards and per-model code ownership, and unified Jira, Jenkins, and Opsera into a single DevOps workflow.
•
Established documentation as engineering infrastructure: weekly KPI decks, Sphinx API docs, onboarding guides, and the HSSLab Web User Guide.
Designed and built the HSSLab platform from 0→1, the internal engineering platform the Staff role above now owns and scales.
•
Designed and built a Django web application with Celery + Redis for background job scheduling and reliable async processing of long-running simulation workloads.
•
Designed the initial Channel Library data model and search UX to store, query, and filter channel artifacts efficiently.
•
Integrated Papermill to execute parameterized .ipynb pipelines for automated reporting and batch analysis.
•
Scaled the Celery + Redis worker pool from 5 to 25 as simulation demand grew.
•
Implemented CI/CD with GitHub Actions to automate deployment and enforce a consistent release workflow on every merge.
•
Established Git branching strategy and PR review standards to improve code quality and team velocity.
•
Developed internal dashboards and reporting utilities using Matplotlib to surface key metrics and experiment results.
•
Evolved storage across MongoDB and PostgreSQL to improve persistence and query performance as access patterns changed.
Teaching Assistant - CSE Department
Santa Clara University
September 2023 – June 2024
•
Santa Clara, CA
•
Taught Introduction to Programming Languages.
•
Instructed Data Structures and Algorithms.
•
Led Operating Systems lab sessions.
•
Mentored students through debugging, assignments, and project milestones.
•
Supported course operations: office hours, grading support, and clarifying concepts with practical examples.
Led research and development for SMAR (Systematic Mobile Application Reviews) tool.
•
Led development for an HCI research tool end-to-end: requirements, UX iteration, implementation, and deployment.
•
Built a production-grade web application using React, Material UI, and Node.js.
•
Owned cloud deployment and operationalization using AWS (EC2, S3, Route 53) and Docker containerization.
•
Built/maintained data acquisition workflows (Play Store scraping) to support research and experimentation.
•
Conducted user studies and incorporated findings into iterative product improvements.
•
Coordinated a multi-person research/development team and aligned deliverables to lab goals and timelines.
•
Built a Python MVC application using Django with GraphQL integration and a Bootstrap-based UI.
•
Improved CI/CD workflows using Gerrit and Jenkins, reducing integration time by ~20%.
•
Collaborated with engineers to ship models through code review and automated pipelines.
•
Focused on maintainable structure, predictable deployments, and developer experience improvements.
•
Built an enterprise threat detection and response solution for SIEM workflows, improving security posture.
•
Created firewall log mapping and parsers using Logstash + Kibana, improving log analysis efficiency by ~30%.
•
Prototyped 30+ UI features using React and Redux to improve scalability and user experience.
•
Designed dashboards and widgets for Events, Signals, and MITRE ATT&CK tactics/techniques mapping.
•
Partnered with UX and cross-functional teams to translate feedback into actionable product improvements.
•
Mentored junior engineers and drove code reviews and quality practices across the team.
•
Architected a Digital Card Facilitator / Click-to-Pay solution at merchant checkout scale, emphasizing data integrity and reliability.
•
Implemented microservices using Java Spring Boot, improving scalability by ~50%.
•
Optimized SQL queries and added indexing to remove bottlenecks, improving retrieval performance by ~40%.
•
Built a secure JavaScript (ES6) SDK using the façade pattern with robust controllers and stronger security guarantees.
•
Integrated SonarQube and raised code quality across security, reliability, and maintainability (~70% improvement).
•
Created and maintained design documentation via Storybook to streamline cross-team collaboration.
•
Developed a Flask + MongoDB microservice to support a risk prediction model.
•
Performed ML time-series forecasting for ATM transaction frequency (~97% accuracy) and surfaced insights for stakeholders.
•
Delivered reporting via Power BI dashboards to visualize trends and operational metrics.
•
Engineered custom Alexa Skills for car and realtor assistants.
•
Used AWS Lambda, DynamoDB, and S3 for robust data handling and storage.
•
Improved performance and reliability through efficient AWS service integration.
•
Worked in a fast-iteration environment to validate ideas and ship usable prototypes.
Projects
Retrieval Engine
Hand-Built Vector Search for an Agentic System
In progress: building and rigorously evaluating a retrieval system for an agent, implementing the index myself (IVF, HNSW, product quantization) rather than importing a vector DB, with an eval harness that scores chunking strategies on nDCG and MRR.
•
In progress: index layer (flat/exact kNN, IVF, HNSW, product quantization) implemented from scratch and benchmarked against FAISS
•
Eval harness scores chunking strategies with nDCG@k and MRR against a hand-labeled query set
•
LangGraph agent layer with citation-grounded output: every claim must cite a retrieved chunk, or the agent flags what evidence is missing
•
Concurrent serving API with P50/P95/P99 latency and recall@k tracking planned for a later milestone
SMAR
Systematic Mobile Application Reviews
HCI research tool that helps researchers analyze mobile app reviews at scale, delivering structured insights from Play Store data.
•
$174,555 National Science Foundation (NSF) award supporting the research.
•
Led development as HCI Research Team Lead at Santa Clara University
•
Built full-stack web application with React, MUI, and Node.js
•
Deployed on AWS (EC2, S3, Route 53) with Docker
•
Integrated Play Store scraping and data analysis pipelines
2248 Linko
Number Merge Puzzle Game
Independently shipped mobile game on the App Store and Google Play. Draw chains of tiles to merge numbers and chase the high score, with Linko the Shiba Inu reacting to every move.
•
Designed, built, and shipped solo, from architecture to App Store / Google Play release
•
Built with Expo / React Native for cross-platform iOS & Android
•
Supabase backend for real-time leaderboards and score sync
•
Offline-first: plays without internet, syncs on reconnect
•
Daily missions, rewarded ads, and in-app purchases
Eatopia
AI-Powered Meal Planning Assistant
Cal Hacks 10 project featuring "Avo," an intelligent chatbot that generates personalized meal plans and recipes using AI.
•
Built mobile app with React Native and TypeScript for cross-platform experience
•
Integrated TogetherAI for intelligent meal recommendations and conversational AI
•
Firebase backend for real-time data sync and user authentication
•
Won recognition at Cal Hacks 10 hackathon
Lets not wait!
Smart Wait Time Optimization
Hackathon-winning platform that reduced wait times using real-time data and intelligent scheduling algorithms.
•
Full-stack application with Python/Django backend and React frontend
•
MySQL database optimization for handling high-volume queries
•
Google Maps API integration for location-based services
•
Won hackathon for innovative approach to reducing service wait times
Contact
Send a Message
Typically replies within 24–48 hours
Connect
Prefer email? Send a note or connect on LinkedIn.
I'm always open to discussing new projects, creative ideas, or opportunities to collaborate on meaningful work.