UC Berkeley · Computer Science

Uriel
Marinez-Uribe.

Software engineering. Applied AI. Systems.

Building software across identity verification, AI-powered quote management, research interfaces, and real-time collaboration.

Explore experience View résumé

Engineering across products and systems.

My work spans full-stack platforms, document intelligence, and real-time systems. At DAISI, I built multi-tenant quote management software and pipelines that turn fixture schedules into validated line items.

My experience also includes telemetry analysis at Socure, genomic workflows for Certis Oncology, and the Advanced Berkeley Comfort research interface at UC Berkeley CEDR.

Colorful blocks being placed and stacked in DriftRoom, followed by a camera sweep around the 3D scene.

Work with measurable outcomes.

2026 / Remote, CA

Socure

Digital Intelligence & Document Verification Intern

Developed data workflows to evaluate telemetry patterns supporting injection-attack detection in document verification.
Implemented signal analysis pipelines for identity verification products, improving fraud investigation and model evaluation workflows.

2026 / Berkeley, CA

DAISI

Co-Founder

Built DAISI's multi-tenant Next.js/TypeScript SaaS on Supabase Auth/Postgres/Storage, with tenant-scoped server actions and typed business logic for quote management.
Engineered a provider-agnostic LLM/VLM pipeline (OpenAI, Anthropic, Ollama) that locates fixture schedules in PDFs, transcribes table crops, and validates line items against Zod schemas to generate quotes.
Fine-tuned Qwen3.5-4B vision LoRA adapters with MLX on a synthetic table corpus, and built a benchmark harness that scores checkpoints and providers against human-reviewed real schedules.

2025 / Remote

Certis Oncology

Software Engineer via Web Development @ Berkeley

Built a NestJS/Angular genomic platform on AWS (EC2, ECS, S3) for FASTQ processing and real-time oncology predictions via an integrated AI pipeline.
Engineered S3 multipart upload orchestration with presigned chunk uploads, retry/abort handling, and pipeline submission logic, increasing reliability for file ingestion and downstream pipeline execution by 40%.
Strengthened platform security with PostgreSQL RLS, RBAC route guards, and unique run-name validation for multi-tenant workflows.

2025 / Berkeley, CA

UC Berkeley CEDR

Software Developer & Researcher (ABC Model Interface)

Shipped features for the Advanced Berkeley Comfort interface, a Next.js/React app used by 50+ researchers, with REST API integration and Three.js visualizations.
Built a clothing-preset workflow with create/edit/import/export flows, localStorage persistence, cross-component state synchronization, and UI validation.

From collaborative systems to inference.

Real-time collaboration

2026

DriftRoom

Real-Time Collaborative 3D Mood Board

Berkeley, CA

Built a full-stack real-time app where users join shareable rooms, see live cursors in a persistent 3D scene, and add shapes/images backed by 7-day Redis room persistence.
Designed a horizontally scalable event pipeline with WebSockets plus Redis pub/sub, presence heartbeats, room lobby assignment, and 20 Hz cursor updates with client interpolation.

WebSocketsRedis3D

Visit DriftRoom

Performance engineering

2026

Llama2.c

Llama-2 Inference on Apple Silicon

Berkeley, CA

Optimized Llama-2 inference in C on an Apple M1 Max to 135 tok/s (14.3× baseline) through compiler optimization and OpenMP; benchmarked thread scaling, scheduling strategies, and ARM NEON against Clang auto-vectorization.

COpenMPARM NEONApple M1 Max

View on GitHub

Skills

Programming languages

PythonJavaScriptTypeScriptC++JavaSwiftLua

Frameworks & libraries

ReactReact NativeNext.jsNode.jsExpress.jsAngularNestJSTailwindCSS

Databases & infrastructure

PostgreSQLAWS (EC2, ECS, S3)FirebaseSentryInfluxDBSupabaseRedisDocker

Tools

GitLinux/UnixPostmanFigma

Spoken languagesEnglish · Spanish

Get in touch.