Nima Jafari
Resume

Software · AI · Machine learning · Robotics

Nima
Jafari.

I’m a software engineer in Austin, Texas. My work spans web and mobile applications, AI systems, machine learning research, and robotics.

Explore my work Let's talk
Nima Jafari
Nima JafariAustin, Texas Research & product engineering
SoftwareAI engineeringML & researchRobotics
Portrait of Nima Jafari

Austin, Texas

About me

I build software across machine learning, robotics simulation, AI systems, and full-stack product development.

My experience includes grant-supported machine learning projects, a pre-seed startup role, and end-to-end product engineering. I’m excited to build new systems with mission-oriented teams.

Education

B.S. Computer Science

California State University, Northridge
May 2026 3.98 / 4.00 GPA

Honors, leadership & life outside code
  • CSUN Computer Science & Engineering Department Scholarships (Guerrera Endowed 2023; Engineering Merit 2024; Trustee Steven G. Stepanek Endowed 2025).
  • Iranian American Women’s Foundation (IAWF) scholarship awardee (2024, 2025); active mentee and volunteer.
  • Dean’s List each semester from Fall 2022 through graduation in May 2026.

Alpha Lambda Delta Honor Society (Co‑founder)

Treasurer (2024): maintained financial records, budget reports, and funding requests to guide leadership decisions.

CSUN Hiking Club (Co‑founder)

Organized weekend hikes; grew membership to ~40 active participants.

Community Engagement

Volunteer with American Red Cross (LA fire) + local orgs; coordinated supply drops and event staffing; tennis team member; music/worship band participation.

Instagram Spotify USTA
Tools I work with

Languages & Frameworks

Python · Java · C++ · JavaScript · TypeScript · Ruby · Rust · HTML/CSS · SQL · PyTorch · TensorFlow/Keras · scikit‑learn · NumPy · Pandas · Jupyter · FastAPI · Node.js/Express · Next.js · React · React Native · Tailwind CSS · Electron · Playwright · Swift · SwiftUI · Solid.js

Machine Learning & Research

Reinforcement Learning (PPO, A2C, DQN, SAC, TD3) · Computer Vision (ViT, ResNet, EfficientNet, Faster R-CNN, YOLO, Swin Transformer) · Vision libraries & tools (OpenCV, torchvision, Detectron2, albumentations) · Transformers & Foundation Models (Hugging Face Transformers, BERT, ViT) · Self-supervised learning (SimCLR, BYOL) · Experiment tracking (Weights & Biases, MLflow) · Model evaluation (mAP, IoU, FID, LPIPS) · Vector search & retrieval · Prompt & tool orchestration · Deployment patterns (TorchServe, BentoML) · Distributed training (DeepSpeed, PyTorch Lightning, Horovod) · Simulation & environments (Gym, MuJoCo, PyBullet, Brax) · Reproducibility & Responsible AI · Supervised learning · Time-series forecasting · Graph-grounded retrieval · LLM application pipelines · Prompt engineering · LLM output validation · AI-generated code review & evaluation · LangChain · Ollama

Databases & Infrastructure

PostgreSQL · MySQL · SQLite · Neo4j · Redis · MinIO · Docker · Docker Compose · Alembic · REST APIs · AWS (RDS, EC2, S3) · Azure · Qdrant · Kubernetes · Data pipelines · Microservices · Cloud deployment

Engineering & Tooling

Git · pytest · CI/CD (GitHub Actions) · Linux + Bash · Figma · Observability (logs/metrics) · Security basics (auth, OWASP) · Performance profiling · Testing strategy · API design · Data structures & algorithms · System design · Debugging · Documentation · GitHub · Cursor · Claude Code · OpenAI Codex · Assisted prototyping · Code review

Data & Analysis

Statistics · Feature engineering · Data cleaning · Exploratory Data Analysis (EDA) · Visualization · Forecasting · A/B testing basics

Experience

Jan 2023 - Present

Contract Projects

Machine Learning Engineer

DoW- and NSF-supported projects, including DARPA and AFOSR programs; appointment administered through University Corporation at CSUN.

Designed the kernel for the Wiring Diagram SDK. Used a Franka Panda environment in NVIDIA Isaac Sim to validate concepts and test the kernel through planar pushing and guarded peg insertion, with structured observations, bounded execution, and auditable trial logs.

More about this role
  • Tested the SDK concepts in an Isaac Sim study comparing wiring-diagram planning with a symbolic baseline: 29/30 original nominal trials reached target, and all 60/60 missing or stale observation fault trials stopped safely.
  • Created reproducible benchmark tooling with held-out task families, ordered trial records, raw evidence, and configuration hashes; results are reported within the frozen study scope.
  • Developed a from-scratch multinomial linear next-action policy baseline with separate training, validation, and evaluation families; qualified legal proposals and safe-stop behavior without direct robot-dispatch authority.
  • Developed Hasse-diagram clustering for sequential reinforcement-learning behavior, recovering two winning strategies across all 125 successful episodes and evaluating robustness against 10% data corruption.
  • Separately trained and evaluated CNN-based figure-skating video classifiers, including transfer learning with a Kinetics-pretrained R(2+1)D-18, and tested pose models including YOLOv8-Pose, MediaPipe, Keypoint R-CNN, and OpenPose.
  • Co-authored two arXiv preprints during this appointment and presented findings at CSUN, Stanford SRC 2026, and the AFOSR 2025 Computational Cognition & Machine Intelligence Program Review.
Aug 2025 - Present

Driftless

Founding AI Engineer

Equity-based, part-time role at a pre-seed startup

Helped build a multi-tenant SaaS product for human teams and AI agents to coordinate work through shared task management and agent execution workflows.

More about this role
  • Delivered features across the React client, Node.js/Express services, MongoDB data layer, and command-line agent tools, including task workflows, agent sessions, and organization-scoped access.
  • Built interactive terminal experiences for agent runs with live task status, inbox handoffs, model/provider selection, onboarding, and resilient process lifecycle handling.
May 2026

SideShift

Full-Stack iOS Engineer (Contract)

New York City, NY

Shipped SwiftUI interface fixes and backend-supported web features; improved data access and caching to reduce unnecessary cloud database and API usage.

More about this role
  • Collaborated on product refinements related to subscription conversion and premium feature adoption.
Jun 2020 - Jun 2021

MISAN Robotic Foundation

Python Software Developer

Remote

Supported firefighting and emergency-response robotics through software development, hardware assembly, mechatronic integration, and prototype testing.

Selected projects

Full-stack · Finance & RAG

PocketPilot

Repo Open app

Personal finance apps for web and mobile, with transactions, budgets, savings goals, and a Rust RAG service. Qdrant combines semantic and keyword retrieval over user-specific financial records, with local inference through Ollama.

  • React
  • React Native
  • TypeScript
  • Rust
  • Firebase
  • Qdrant
  • Ollama
Full stack
React web and Expo mobile clients share domain models and services, backed by Firebase and Rust HTTP services.
Retrieval
Dense embeddings and sparse keyword vectors are stored in Qdrant. Merchant names, dates, and amounts receive explicit keyword boosts; unchanged records are skipped during re-indexing.

Full-stack · Web & mobile

CareerLift

Repo

Designed and built a full-stack AI career platform spanning React web and React Native mobile experiences, with backend systems for resume parsing, job discovery, ATS-style matching, and mock interview coaching.

  • React
  • React Native
  • TypeScript
  • FastAPI
  • Neo4j
  • LangChain
  • Electron
The problem
Bring job discovery, resume matching, and interview preparation into a connected web and mobile experience.
The system
React web and React Native mobile clients supported by backend services for parsing, matching, and coaching. The repository documents the implementation.

Full-stack · Media

ReelMeListing

Repo

Built a browser-based studio that turns property photos into reviewed vertical listing reels, with image-quality checks, optional AI image edits, and video-generation workflows.

  • React
  • FastAPI
  • Python
  • Computer Vision
  • FFmpeg
  • Video Generation
Studio
React and FastAPI handle photo uploads, job records, quality reports, and vertical reel assembly with FFmpeg.
Generated media
Optional image-editing and ComfyUI/LTX workflows produce candidates for human review. Video generation requires a separately configured CUDA workflow; the browser handles planning and deterministic assembly.

Developer tooling

GhostD

Repo

A local TypeScript runtime that records developer-agent events in SQLite and compiles versioned context for side questions to Codex, Claude, or Gemini. Includes a CLI, a VS Code extension, and a read-only MCP server.

  • TypeScript
  • Node.js
  • SQLite
  • MCP
  • VS Code
Context
Captured lifecycle events are normalized into a local SQLite ledger. Compiled context retains source-event references and a specific revision.
Interfaces
Terminal commands, a VS Code extension, and a read-only MCP server expose that context without modifying the original agent conversation.

Clustering + RL

Hasse Clustering

Repo Live demo Paper

A Python engine and browser application for clustering ordered event sequences by Hasse-DAG patterns. The browser runs the engine in a Pyodide Web Worker, with CSV/JSON input, graph visualization, cluster exports, and cancellation.

  • Python
  • Pyodide
  • Web Workers
  • SVG
  • Graph Clustering
Engine
Extracts partial-order patterns from event sequences, including repeated events, and groups sequences by the patterns they support.
Browser app
Runs the canonical Python engine in a Web Worker through Pyodide. Input and computation stay in the browser; the deployment serves static assets.
More projects 5 projects

SDK · Kernel design

Wiring Diagram SDK

Related experience

Designed the kernel for the Wiring Diagram SDK, using NVIDIA Isaac Sim as a test environment for concept validation and evaluation.

  • Python
  • Kernel design
  • Wiring diagrams
  • NVIDIA Isaac Sim
Kernel design & validation
Designed the kernel for the Wiring Diagram SDK. Used a Franka Panda environment in NVIDIA Isaac Sim to validate concepts and test the kernel through planar pushing and guarded peg insertion, with structured observations, bounded execution, and auditable trial logs.
Evidence, within scope
Tested the SDK concepts in an Isaac Sim study comparing wiring-diagram planning with a symbolic baseline: 29/30 original nominal trials reached target, and all 60/60 missing or stale observation fault trials stopped safely.
Reproducibility
Created reproducible benchmark tooling with held-out task families, ordered trial records, raw evidence, and configuration hashes; results are reported within the frozen study scope.

Research · Reinforcement learning

StrategyMining

Repo Preprint

Designed two versions of a video game, trained PPO agents to play them, and analyzed gameplay traces with Hasse-diagram clustering to recover the winning strategies.

  • PPO
  • PyTorch
  • Hasse diagrams
The question
How can successful agent behavior be organized into understandable strategies rather than a collection of opaque traces?
My contribution
Developed Hasse-diagram clustering for sequential reinforcement-learning behavior, recovering two winning strategies across all 125 successful episodes and evaluating robustness against 10% data corruption.
Method
Custom game environments, trained PPO agents, and Hasse-diagram clustering, with DBSCAN and hierarchical clustering comparisons.

AI engineering · AgenticAI

AgenticAI

Repo

An AI-assisted refund-support workflow with a Streamlit chat frontend and FastAPI backend. A deterministic policy engine validates customer and order ownership, evaluates policy, and keeps the LLM limited to missing-information prompts and customer-facing explanations.

  • Python
  • FastAPI
  • Pydantic
  • Ollama
The boundary
A deterministic policy engine makes policy decisions; the language model gathers missing information and explains the outcome.

Automation

JobApplyX

MV3 extension and Node/Express backend with SQLite and Ollama for parsing job descriptions, answering screening questions, and generating cover letter PDFs.

  • JavaScript
  • Node.js
  • Express
  • SQLite
  • Ollama
  • MV3
Repo

Computer Vision

Figure-Skating Jump Analysis Pipeline

Analyzed figure-skating jump videos with CNN classifiers and pose-estimation models, then applied interpretable Hasse-DAG clustering to motion-event sequences.

  • CNN
  • YOLOv8-Pose
  • Computer Vision
  • Hasse Clustering
Repo Paper

Research & publications

Reinforcement learning, motion analysis, and the structure hidden inside sequential data.

arXiv Preprint

From Data to Concepts via Wiring Diagrams (arXiv:2511.20138)

Designed two versions of a video game and trained PPO reinforcement-learning agents to play them. Hasse-clustering algorithms analyzed gameplay traces to recover winning strategies, with comparisons against DBSCAN and hierarchical clustering.

Zenodo DOI: 10.5281/zenodo.17315753

Game Version 3: RL Strategy Mining

Further data paper for V3: dual-strategy discovery and symbolic analysis in the updated RL environment.

Resume

Engineering roles, research, education, and technical skills in one place.

Open resume Download PDF

Contact

mohammadnimajafari@yahoo.com
GitHub