Dev Amin — Software & AI/ML Engineer

Rutgers University–New Brunswick · CS & Data Science

Distributed systems, AI/ML, and a lot of chess.

I build systems that stay correct under pressure. Most recently, Pulse: it reads 140 news sources in 24 languages, groups articles about the same event into live stories across languages, and can replay any past hour from its Kafka log with nothing lost and nothing counted twice.

See the projects Résumé

140 news sources, 24 languages

0 stories lost or duplicated on replay

3.7 GPA, Dean's List

Rust♟︎Java 21♟︎Python♟︎TypeScript♟︎ Kafka♟︎Redis Cluster♟︎PostgreSQL♟︎Spring Boot♟︎ Axum♟︎Tokio♟︎gRPC♟︎WebSockets♟︎ OpenTelemetry♟︎Docker♟︎PyTorch♟︎LangChain♟︎ React♟︎Next.js♟︎ Rust♟︎Java 21♟︎Python♟︎TypeScript♟︎ Kafka♟︎Redis Cluster♟︎PostgreSQL♟︎Spring Boot♟︎ Axum♟︎Tokio♟︎gRPC♟︎WebSockets♟︎ OpenTelemetry♟︎Docker♟︎PyTorch♟︎LangChain♟︎ React♟︎Next.js♟︎

1. The Opening

Between moves

I'm Dev Amin, a Computer Science and Data Science student at Rutgers University–New Brunswick. I work across software engineering and AI/ML, from streaming data pipelines and distributed systems to the machine learning models that run on top of them.

Away from the keyboard I'm usually in the middle of a chess game, playing video games, or replaying a game to find the move I missed.

School: Rutgers University
Degree: B.S. Computer Science & Data Science
Graduating: December 2026
Status: U.S. citizen

2. The Pieces

What I play with

Hover a card to see how its piece moves.

Languages

The queen — goes anywhere

PythonJavaRustTypeScript JavaScriptSQLCRMATLAB

Distributed Systems

The rook — holds the line

Spring BootAxumTokioFastAPI Kafka / RedpandaRedis ClusterPostgreSQL gRPCWebSocketsFlask

AI / ML

The bishop — sees the long diagonal

PyTorchTensorFlowscikit-learnLangChain RAGLLMs (Ollama, GPT-4)EmbeddingsMinHash LSH PandasNumPyPySpark

Infra & Tooling

The knight — gets around obstacles

Docker ComposePrometheusGrafanaOpenTelemetry Jaegerk6ToxiproxyGitHub Actions TestcontainersReactNext.jsTauri

3. Game Record

Experience

1.
AI Engineering Fellow

Headstarter AI · Remote

Jul 2024 — Sep 2024
- Seven-week applied AI fellowship built around one-week sprints that mirrored a production engineering team.
- Weekly hackathons and peer code review — version control discipline, systematic debugging, agile process.
- Optimized ML pipelines and data processing, cutting model inference time by over 25%.
- Prototyped and iterated on features that reached 50 active beta users during testing.
2.
Community Service Officer

Rutgers University · New Brunswick, NJ

Nov 2024 — Mar 2025
- Patrolled a 50,000+ student campus on foot, bicycle, and in marked vehicles, reporting incidents to dispatch.
- 40-hour NJ Public Safety Association certification in emergency response, radio protocols, and de-escalation.
- Managed access control and bag screening at events of 5,000+ attendees alongside sworn officers and EMTs.
- Ran the shuttle and walking escorts for 100+ people nightly.

4. Best Games

Projects

Each one has a walkthrough — open "How it works" to step through the architecture.

Streaming · ML

Pulse

The world's news, grouped into stories as it breaks

Reads 140 sources in 24 languages and clusters articles about the same event into live stories — across languages — then lets you scrub back to any moment in time.

Exactly-once · 5,037 articles replayed in 6.1 s Re-run any past hour and get the identical result — nothing lost, nothing counted twice.

Streaming story clustering with MinHash LSH near-duplicate detection plus embedding similarity, merging coverage across sources in real time.
Hand-rolled exactly-once processing in Rust — checkpointing, watermarks, transactional offset commits — and a replay service that rebuilds state from the log.
Story processor handles ~900 articles/s; spoken story summaries in 71 languages.

Rust · Tokio · Python · Kafka / Redpanda · PostgreSQL · React · TypeScript · Docker

Distributed systems

Surge

Flash-sale ticketing that never oversells

Polyglot microservices built to absorb flash-sale traffic spikes — a waiting room, live seat map, seat holds, and a checkout saga with payments.

10,000 seats · 174 s · 0 oversold Held 0 invariant violations even with Redis primaries killed mid-sale.

Redis Lua seat holds as the fast, lossy layer; Postgres constraints as the source of truth. Checkout is a saga with idempotency keys and a transactional outbox.
A reconciler audits invariants while a chaos controller kills Redis primaries and partitions services.
Next.js live seat map over WebSockets, end-to-end OpenTelemetry tracing, load-tested with k6.

Java 21 · Spring Boot · Rust · Axum · Python · FastAPI · Redis · PostgreSQL · Redpanda · Next.js

Systems · AI

EpochOS

A local-first AI layer for the desktop

Wraps file access, process control, and terminal monitoring in a natural-language interface, running on local models on your own machine.

70% of repetitive workflows automated One React codebase across macOS, Windows, and Linux.

Local-first system layer in Rust + Tauri that turns plain-English requests into OS actions.
AI memory engine on RAG (LangChain + SQLite) indexing workspace files and app state — 35% higher accuracy than stateless prompts.
Function-calling orchestration chains tools across multi-step tasks, cutting completion time by 50%.

Rust · TypeScript · React · Tauri · LangChain · Ollama · SQLite · Docker

5. Your Move

Let's talk.

Open to software engineering and AI/ML internships and new-grad roles. Email is the fastest way to reach me.

Email dev.amin1016@gmail.com Phone (732) 781-8528 GitHub @DAmin05 ↗ LinkedIn /in/dev-amin05 ↗ Résumé View PDF ↗