Hello! 👋
I'm Viken Parikh
AI/ML Software Engineer at Electronic Arts, building LLM agents, MLOps, and generative-AI platform services for game and live-service teams.
6+ years across EA, Microsoft, and PayPal shipping production systems at scale: 50K+ QPS, ~$2B daily volume, 99.99% uptime, 25K+ developers served, ML-driven security across 50K+ repositories.
Vancouver, BC, Canada
Experience
Electronic Arts 
Microsoft 
PayPal 
About Me
AI/ML Software Engineer at Electronic Arts (EA), building the central AI platform that game studios and live-service teams ship on. I work where ML research, MLOps, and high-scale backend meet, the boring stuff that makes the demo actually ship: models you can trust, evals you'd promote on, and serving that holds its SLOs under real traffic. 6+ years across EA, Microsoft, and PayPal designing intelligent platforms, serving 25K+ developers and processing $2B+ daily volume, with deep hands-on experience in LLMs, RAG, agent orchestration, fine-tuning, distributed systems, and cloud security.
AI/ML Software Engineer 2 at Electronic Arts: building EA's central AI platform powering studios such as EA Sports FC, Madden, Apex, Sims, Battlefield, and F1.
6+ years across EA, Microsoft, and PayPal shipping production AI and distributed systems.
Architected agentic systems with LLMs (Claude, GPT, Llama, Mistral): LangGraph + MCP, retrieval-augmented generation, evaluation harnesses, and safety guardrails.
Shipped three production-grade AI products end-to-end (edumind-ai, neuralverse-ai, medmind-ai) on a self-hosted, multi-service AI platform.
Served 25,000+ developers through Microsoft platform and API initiatives.
Enabled $2B+ in daily payment volume at PayPal with 99.99% uptime and 50K+ QPS.
Scanned 50,000+ repositories for code, secret, dependency, and IaC vulnerabilities with ML-driven detection at Microsoft Defender for DevOps.
Improved card-traffic authorization success by ~2–3% via ML-guided routing + retry at PayPal.
Reduced onboarding time by ~50% through feature engineering + A/B experimentation at Azure Cloud Security.
Published IEEE research cited 100+ times.
Deployed YOLO computer-vision pipeline processing 10,000+ images/day at ~95% accuracy.
Skills
Languages
PythonTypeScript / JavaScriptJavaC#GoRust (learning)SQLBashCUDA / Triton kernels (basics)
Foundation Models & Generative AI
Frontier LLMs: Claude (Opus, Sonnet, Haiku), GPT-4/4o/5, Gemini, Llama 3, Mistral, Mixtral, DeepSeek, QwenOpen-weight multimodal: Llava, Idefics, Florence-2, SAM, WhisperImage / video / audio gen: Stable Diffusion / SDXL, ControlNet, Flux, Suno, ElevenLabs, Sora-classCode models: Claude Code, GPT-4-code, CodeLlama, StarCoderPrompt engineering: structured outputs, JSON mode, function/tool calling, constrained decoding (Outlines, Guidance)Context engineering: long-context, prompt caching, prefix sharing, cost-vs-latency tradeoffsDistillation + small-model strategy (TinyLlama, Phi, distilled fine-tunes for edge / on-device)
LLM Agents & RAG
Agent frameworks: LangGraph, LangChain, CrewAI, AutoGen, Anthropic MCP, OpenAI Assistants, Google Agent Builder / Vertex AI Agents, Smol AgentsTool use + ReAct + plan-and-execute + tree-of-thought + reflection patternsMulti-agent orchestration: supervisor + worker, debate, swarm, hierarchicalRetrieval-augmented generation: hybrid (BM25 + dense), rerankers (Cohere, bge-reranker), query rewriting, HyDEVector databases: pgvector, Pinecone, Weaviate, Qdrant, Milvus, FAISS, Chroma, LanceDBEmbeddings: OpenAI, Cohere, Voyage, BGE, Nomic, JinaMemory systems: episodic / semantic / scratchpad, vector + graph hybridsKnowledge graphs: Neo4j, GraphRAG, entity + relation extractionFunction calling, structured I/O, schema-guided generationBrowser / OS agents (Anthropic Computer Use, Operator)
Machine Learning & Deep Learning
Core ML: supervised, unsupervised, semi-supervised, self-supervised, contrastive learningDeep learning architectures: Transformers, CNNs, RNN/LSTM/GRU, GNNs, Diffusion, VAEs, GANs, Mixture-of-ExpertsPyTorch, TensorFlow, JAX, Hugging Face Transformers + TRL + PEFT + AccelerateSupervised fine-tuning (SFT), instruction tuningParameter-efficient: LoRA, QLoRA, DoRA, prefix tuning, adaptersPreference / RL: RLHF, DPO, IPO, KTO, RLAIF, PPODistributed training: DeepSpeed, FSDP, ZeRO, Megatron-LM, tensor + pipeline parallelQuantization: GPTQ, AWQ, GGUF, bitsandbytes, FP8 / INT8 / INT4Classical ML: scikit-learn, XGBoost, LightGBM, CatBoostDeep learning: CNNs, Transformers, GNNs, diffusion, VAEs, recommenders (Two-Tower, DLRM)Reinforcement learning: PPO, A2C/A3C, SAC, DQN, multi-agent RL, MARL for game AITime-series: Prophet, NeuralProphet, TimesFM, classical ARIMA
MLOps & ML Platform
Experiment tracking + model registry: MLflow, Weights & Biases, Comet, NeptuneTraining orchestration: Ray + Ray Train, Kubeflow, Airflow, Prefect, Dagster, SageMaker Pipelines, dstackWorkflow automation: n8n, Zapier (low-code), Temporal (durable workflows)Feature stores: Feast, Tecton, SageMaker Feature StoreInference serving: vLLM, TGI, Triton, BentoML, TorchServe, Ray Serve, SageMaker, BedrockAI observability + tracing: LangSmith, Langfuse, OpenTelemetry, Arize, Helicone, WhyLabsLLM evals: Promptfoo, OpenAI Evals, Inspect, HELM, lm-eval-harness, LLM-as-judgeOnline experimentation: A/B, multi-armed bandits, interleaving, CUPED, regression-discontinuityModel + data versioning: DVC, LakeFS, Hugging Face HubDrift, fairness, calibration monitoring; canary + shadow + rollback strategiesGPU + accelerator economics: autoscaling, spot, multi-instance GPU, request batching
Data Engineering
Streaming: Kafka, Kinesis, Pulsar, FlinkBatch: Spark, Databricks, BeamDataFrames: Pandas, Polars, DuckDB, DaskWarehouses + lakes: Snowflake, BigQuery, Redshift, Iceberg, Delta Lake, ParquetTransformation: dbt, SQLMeshSynthetic data + augmentation, deduplication, dataset curationETL/ELT, CDC (Debezium), schema evolution
Cloud & Infrastructure
AWS: SageMaker, Bedrock, Inferentia/Trainium, ECS, EKS, Lambda, S3, DynamoDB, Step Functions, Kinesis, MSK, OpenSearchAzure: Azure ML, OpenAI Service, AKS, FunctionsGCP: Vertex AI, GKE, Cloud Run, BigQueryKubernetes, Helm, Argo, KEDADocker, ContainerdIaC: Terraform, Pulumi, CDKEdge + reverse proxy: Caddy, Nginx, EnvoyCloudflare: Workers, Tunnel, Access, R2, KV, AI GatewayObservability: Prometheus, Grafana, Datadog, OpenTelemetry, ELK
Backend & Frameworks
FastAPI, Flask, Django, LitestarNode.js, Express, Hono, Next.js (App Router), React, React Native + Expo, SvelteKitSpring Boot, ASP.NET CoregRPC, GraphQL, REST + OpenAPIPydantic, Zod, Protobuf schemasWebSockets, Server-Sent Events, real-time streaming for LLM token streams
Databases
PostgreSQL (+ pgvector, TimescaleDB, PostGIS)MySQLRedis, ValkeyMongoDB, DynamoDB, CassandraClickHouse, DuckDBElasticsearch, OpenSearch, MeilisearchVector: Pinecone, Weaviate, Qdrant, Milvus, FAISS, Chroma
Safety, Evals & Responsible AI
Red-teaming, jailbreak + prompt-injection defenseContent safety filters, PII redaction, toxicity + bias detectionHallucination grounding + citation enforcementDifferential privacy, federated learning (basics)Model + data cards, provenance, audit trailsEval pipelines: golden sets, LLM-as-judge calibration, human-in-the-loop review
Tools & Practices
Git, GitHub Actions, GitLab CI, ArgoCDTDD, integration tests, eval-gated promotion, regression CI for MLLinters + formatters: ruff, mypy, eslint, prettierProfiling: py-spy, scalene, NVIDIA Nsight, PyTorch profilerNotebooks: Jupyter, Marimo, ColabDesign: Figma; Diagramming: Mermaid, ExcalidrawAgile, OKRs, system-design reviews, on-call + incident-response
Experience
AI/ML Software Engineer 2
Electronic Arts (EA)- Building EA's central AI platform: shipping reusable machine-learning, deep-learning, and generative-AI services, agent frameworks, and a composable skills / tools layer that game studios and live-service teams (EA Sports FC, Madden, Apex Legends, The Sims, Battlefield, F1) plug into for personalization, content generation, NPC behavior, and player-experience features.
- Owning ML lifecycle infrastructure end-to-end: Python (PyTorch + Hugging Face) training stacks, distributed fine-tuning (LoRA/QLoRA/SFT/DPO), feature pipelines, experiment tracking (MLflow / W&B), model registry, low-latency inference serving (vLLM / TGI / Triton / SageMaker), drift + cost + quality observability: with hard SLOs on latency, reliability, and per-request economics.
- Designing agentic systems on top of LLMs (GPT-class, Claude, Llama, Mistral, in-house fine-tunes): retrieval-augmented generation over EA knowledge corpora, structured tool use, multi-agent orchestration (LangGraph / MCP), eval harnesses (LLM-as-judge + golden-set regressions), and guardrails (prompt-injection defense, PII handling, content safety) so AI features ship measurably and safely at game-scale traffic.
- Architecting platform services on AWS (SageMaker, Bedrock, ECS/EKS, Lambda, S3, DynamoDB, Step Functions) + IaC (Terraform): multi-tenant inference gateways, prompt + model routing, caching, rate limits, audit logging: that other EA teams consume via golden-path SDKs and APIs.
- Driving AI adoption across EA: partner with studios: including EA Sports FC: on use-case discovery (player matchmaking, recommendations, generative content, anti-cheat, live-ops personalization, NPC behavior, in-game support, dynamic difficulty), publish reference patterns + internal docs, run enablement and design reviews, and measure adoption + impact (active integrations, eval-gated launches, incident rate, dollar value of compute saved).
- Stack: Python, PyTorch / TensorFlow (deep learning + classic ML), Hugging Face, LangChain, LangGraph, MCP, Google Agent Builder / Vertex AI, n8n + Airflow + dstack for orchestration, FastAPI, Ray, Kubernetes, Docker, Terraform, AWS (SageMaker / Bedrock), Postgres + pgvector, Redis, Kafka.
Independent AI/ML Engineer & Builder
Self-Directed AI Products- Shipped three production-grade AI products end-to-end: edumind-ai (adaptive learning + analytics), neuralverse-ai (multi-agent platform + knowledge graphs), medmind-ai (clinical-decision support): from problem framing and UX through deployment, observability, and real-user feedback.
- Architected and operated a self-hosted, multi-service AI platform powering all of the above: Postgres + pgvector, FastAPI + Next.js services, Docker / Caddy / Cloudflare edge, on-VPS CI/CD, alerting, backups: owning the full MLOps and platform-engineering stack hands-on.
- Built agentic systems on top of Claude / GPT / Llama using LangGraph + MCP: a 15-session orchestrator fleet with deterministic guardrails, retrieval-grounded reasoning, eval harnesses, prompt-injection defenses, and structured tool use.
- Deepened frontier AI/ML skills by building, not consuming: LLM agents, RAG with hybrid retrieval + rerankers, fine-tuning (LoRA / DPO), prompt + structured-output design, LLM-as-judge evaluation, AI observability (LangSmith / OpenTelemetry).
- Outcome: joined Electronic Arts as AI/ML Software Engineer 2 applying this work directly to game and live-service AI platforms.
Software Engineer 2
Microsoft, Seattle & Vancouver- Built ML-driven security analysis tooling for Defender for DevOps, scanning 50K+ Azure DevOps/GitHub repositories to detect code, secret, dependency, and IaC vulnerabilities using intelligent pattern matching, reducing detection time by ~40% and increasing remediation throughput by ~60%.
- Designed and shipped predictive security dashboards using React and Knockout with ML-backed insights, delivering unified code-to-cloud visibility and driving ~30% higher feature adoption among 25K+ developers.
- Developed feature engineering pipelines and A/B testing frameworks for Azure Cloud Security, leveraging data-driven recommendations to streamline onboarding and cut setup time by ~50%.
Software Engineer 2
PayPal, San Jose- Architected and implemented an intelligent payment authorization system using feature engineering, controlled experiments, and ML-guided policies to optimize routing and retry logic, improving transaction efficiency and increasing authorization success by ~2–3% on high-volume global card traffic.
- Built high-throughput tokenization SDKs and APIs using Java (Spring Boot), Couchbase, Docker, and Kafka, supporting 50,000+ QPS and ~$2B daily volume with 99.99% uptime while integrating ML-based fraud detection and experimentation.
- Engineered a secure tokenization platform with intelligent lifecycle management, improving reliability and flexibility for stored payment instruments across multiple PayPal flows while maintaining PCI-compliant practices.
- Mentored four junior engineers on microservices, observability, and experimentation-driven development, establishing best practices for production systems and data-driven architecture.
Software Engineer
Decision Theater Network, Arizona- Developed an ML-assisted visualization and simulation platform (Python, JavaScript) for over 50 research projects, reducing analysis cycles by ~45% and enabling faster insight generation.
Education
Master of Computer Science (Data Science and AI)
Arizona State University, Tempe, AZ
- Coursework: Statistical Machine Learning, Artificial Intelligence, Multi-Robot Systems, Semantic Web Mining, and Cloud Computing.
Bachelor of Technology, Computer Engineering
Mumbai University, India
- Coursework: Machine Learning, Neural Networks, Fuzzy Logic, AI, Data Mining, and Computer Simulation Modeling.
Projects
Featured AI Projects
AI-powered educational intelligence platform with adaptive learning and analytics.
Unified AI development platform with multi-agent systems and knowledge graphs.
AI-powered medical intelligence platform with clinical decision support.
Other Projects
Cloud platform for scalable object detection using YOLO and AWS services.
AI-driven financial intelligence platform with portfolio optimization and risk analysis.
Music and text generation using deep learning models.
Photo captioning application powered by neural networks.
Website about various travel places in India with backend AI features.
Movies and TV shows recommendation engine with backend AI.
Personal travel app with ML-powered recommendations and sentiment analysis.
Portal connecting donors with NGOs to reduce food waste.
Prediction of house prices based on various attributes.
View all projects and source code on github.com/vikenparikh
Contact Me
Or email me directly at vsparikh1996@gmail.com