Berkeley EECS · Open to SWE internships

Daniel Firoozabadi

I build systems that make sense of messy data at scale — agentic pipelines, hybrid retrieval, and backend infrastructure that holds up under real load.

See the work Email GitHub LinkedIn
Daniel Firoozabadi
Berkeley, CAEECS '28

What I build

Hybrid retrieval, end to end

The hard part of retrieval isn't the model — it's getting a trustworthy answer out of six disagreeing systems fast enough that someone will actually use it. Graph traversal finds the structural path; vector search finds the semantic one; the merge step reconciles them. Here's the shape of it.

Query · 6 sources · graph + vector merge

Sources queried

Resolved in—

A stylised illustration of the retrieval architecture I built at Capital One, not live infrastructure. Fire a query and watch the fan-out, merge, and ranking.

Work

Where I've worked

Capital One

Software Engineering Intern

Jun 2026 — Present Now

  • LangGraph
  • Chroma
  • PyArrow
  • Kubernetes
  • Cut a 4-week manual Excel reconciliation process to under 2 days by building an agentic RAG system (LangGraph, Chroma, PyArrow) automating GL reconciliation between Hyperion RACR and Workday FDM for a $10B-spend ledger migration.
  • Returned discrepancy lookups in under 3 seconds across 6,500+ SOX accounts via a hybrid retrieval pipeline combining graph traversal and semantic vector search across 6 data sources, using PyArrow predicate pushdown on 6M+ monthly fact rows.
  • Enabled 20+ Finance analysts to self-serve Balance Sheet and Income Statement reconciliation without HFM/Workday expert support by deploying a containerized ReAct agent (Docker, Kubernetes) with an 8-tool toolkit and Gradio interface.

Olfera

Software Engineering Intern

May 2025 — Aug 2025

  • MySQL
  • GraphQL
  • AppSync
  • Reduced redundant queries by ~40% and improved data integrity by designing 7+ normalized MySQL schemas for users, projects, experiments, and SOPs.
  • Developed secure GraphQL/AppSync API endpoints with role-based access control (RBAC) to protect sensitive research data and support scalable access patterns.

Runopt

Software Engineering Intern

Jan 2025 — May 2025

  • PyVista
  • Plotly
  • OpenMDAO
  • Developed a Python 3D visualization framework with PyVista and Plotly, enabling engineers to generate stormwater detention layouts in under 10 seconds per simulation.
  • Implemented a knapsack-based optimization pipeline across 9+ underground detention models and integrated OpenMDAO, reducing cross-module runtime by ~30%.

Projects

Things I've shipped

Tandem

TypeScript · React · Node.js · Anthropic SDK · WebSockets

  • A real-time collaborative coding environment where multiple developers share, branch, and replay live AI sessions simultaneously.

Final partner interview, a16z Speedrun · Pitched to Xfund & Dorm Room Fund

ShopSync

Next.js · Supabase · Vercel · OpenAI API

  • A full-stack LLM platform that extracts structured inventory data to power personalized product matches for shoppers and listing suggestions for businesses.
  • Ingestion and normalization pipelines for scraped and API-fed data, with embeddings-based retrieval in Supabase.

Falling Walls Lab SF Finalist

View repo

Agentic Penetration Testing Framework

Python · AutoGen · PentAGI · Nmap · Nuclei

  • Led architecture and security-framework planning, then built an end-to-end AI-assisted pentesting pipeline in weekly collaboration with a Senior InfoSec Engineer at a higher-ed institution's InfoSec team.
  • Designed a multi-agent system coordinating reconnaissance, vulnerability classification, and report generation with human-approval gates, append-only audit logging, and an emergency kill switch; confirmed 5 valid vulnerabilities from ~10 findings.

5 valid vulnerabilities confirmed from ~10 findings

Private repo — code involves a partner institution's infrastructure

Goldman Sachs Mutual Fund Challenge

Java · Spring Boot · Angular · OpenAI API

  • Built for the Goldman Sachs Emerging Leaders 2026 challenge — a portfolio projection tool combining CAPM expected-return modeling with LLM-generated allocation.
  • Connected the Angular frontend to the Spring Boot backend's future-value projection endpoint as part of a 5-person team.

Goldman Sachs Emerging Leaders 2026 · team of 5

View repo (team project)

Education

UC Berkeley

B.S. Electrical Engineering & Computer Sciences · Aug 2024 – May 2028 · GPA 3.5

Data Structures & Algorithms · Computer Architecture · Optimization Models · Discrete Math & Probability

Selected for

Goldman Sachs Emerging Leaders Series — Software Engineering Track, Dec 2025 – May 2026. Built quantitative analytics infrastructure for a mutual fund forecasting platform.

Y Combinator Startup School 2026 — hand-selected, flown to San Francisco for two days among top builders worldwide.

Ascend Berkeley (Tech Chair) · AI Entrepreneurs at Berkeley · IEEE · 2× SIG Discovery Day · Point72 Academy Spring Sessions

Stack

Languages

Python, Java, C/C++, SQL, JavaScript, TypeScript

Frameworks & ML

PyTorch, TensorFlow, NumPy, Pandas, React, Node.js, LangChain, AutoGen

Tools

AWS, Docker, Kubernetes, Git, MySQL, Vercel, Supabase, CI/CD, REST APIs, Terraform

Contact

dfiroozabadi@berkeley.edu
Berkeley, CA GitHub · LinkedIn