Seif Ali
Builds AI systems that run where the data lives.
Final-year Computer Science (AI & Big Data) at the University of Wollongong. Grounded RAG pipelines, local-first tools, and full-stack applications — engineered to cite their sources, not guess.
Profile
Most AI demos guess. Mine cite their sources.
I'm a final-year Computer Science student at the University of Wollongong, and the thread through everything I build is grounding — AI that answers from real documents, real manuals, real code, and shows you exactly where the answer came from. That means retrieval pipelines with page-level citations, local-first tools where sensitive data never leaves the machine, and systems that refuse to make things up when the evidence isn't there.
I started the degree at UOW Dubai in 2023 and transferred to Australia in 2025 — chasing bigger opportunities and colder weather. When I'm not coding, I'm deep in a game or stress-testing the newest AI tools.
- Location
- Sydney, NSW — Australia
- Status
- Final year · graduating Dec 2026
- Degree
- BCompSci — AI & Big Data, UOW
- Focus
- Grounded AI · RAG · full-stack systems
- Languages
- English (fluent) · Arabic (native)
- Currently
- Capstone: NLP × program analysis
Selected Work
Sydney Tech Pulse↗
Live job-market data pipeline
Data EngDaily ELT pipeline that mines Sydney job postings for the skills employers actually hire for, loading Postgres views behind a live trends dashboard.
90-pattern skill taxonomy · $0/mo to run
- /Python
- /PostgreSQL
- /Docker
- /GitHub Actions
- /Supabase
ZynTrace
Software update tracing — capstone
NLP × SecDiffs code changes against release notes to surface silently-patched vulnerabilities and cloned defects that update descriptions never mention.
Program analysis × transformer NLP
- /Python
- /NLP
- /Program Analysis
- /Transformers
Command Center↗
Productivity dashboard
Full-StackEverything-dashboard with Notion sync, drag-and-drop layout, and a Ctrl+K command palette — timetable, workouts, prayer times, Pomodoro.
16 widgets · Notion API sync
- /Next.js
- /TypeScript
- /Zustand
- /Notion API
TaskFlow↗
Full-stack kanban
Full-StackKanban task manager with drag-and-drop, priority labels, multi-project boards, and JWT authentication.
Auth + DB + deploy, end to end
- /Next.js
- /Supabase
- /Prisma
- /NextAuth.js
Talent Matching Platform
Team project — backend
BackendJob–candidate matching engine built by a five-person team. Owned the backend: weighted recommendation scoring and fuzzy search.
55-point scoring · CI on every push
- /PHP 8
- /MySQL
- /PHPUnit
- /GitHub Actions
Archive
RAG PDF Chat
AI / RAGUpload any PDF, ask questions, get cited answers from a locally-hosted LLM. No API keys, no data leaves the machine.
- /Python
- /LangChain
- /FAISS
- /Ollama
Big Data ML
MLDecision-tree classifiers written from scratch (information gain + Gini) with weighted ensembling, applied to UCI datasets.
- /Python
- /pandas
- /NumPy
ai-commit
Dev ToolsGenerates conventional git commit messages from staged diffs using a local Ollama LLM.
- /Python
- /Click
- /Ollama
Text Summariser
AISummarises any text with locally-running LLMs — streaming output, three summary styles, zero API keys.
- /Python
- /Streamlit
- /Ollama
Case File — GEARHEAD
Workshop manuals are 600-page PDFs nobody has time to search — so garages guess, and drivers pay for parts they didn't need. GEARHEAD turns the manual for your exact car into a diagnostic interview that answers with page-level citations and the manual's own diagrams. Built GCC-first: starting with the Nissan Patrol, the region's workhorse.
Pipeline — Fig. 01
Architecture
- Ingestion
- PyMuPDF + custom section-aware chunker
- Vector store
- ChromaDB — 2,450 embedded chunks
- Inference
- Claude API or local Ollama, switched with one env var
- API
- FastAPI
- Grounding
- Page-level citations + the manual's actual diagrams
How it works
Describe the fault
A symptom in plain words or an OBD-II code — “P0420 after cold start” is enough.
Get interviewed
Follow-up questions like a mechanic's intake: when, how often, what changed recently.
Retrieval
Hybrid search over the indexed workshop manual for that exact vehicle — not generic car advice.
Cited fix
Diagnosis with the manual page reference and the actual wiring or component diagram.
$ gearhead diagnose
vehicle: Nissan Patrol Y61 4.8
fault: P0420 — only on cold mornings
… retrieving from service manual (2,450 chunks)
Likely an aging upstream O2 sensor skewing readings before warm-up — test its voltage response before touching the catalytic converter. Replacing the cat on this code without sensor data is the expensive mistake.
— Engine Control section, p. EC-247, Fig. 12
0
Workshop manuals indexed
0
Retrievable chunks
0%
Works offline — local LLM
Stack
AI / ML
- RAG
- LangChain
- FAISS
- Transformers
- Multimodal LLMs
- Ollama
- scikit-learn
- TensorFlow
- Spark MLlib
- NLP
- Prompt Engineering
Languages
- Python
- TypeScript
- JavaScript
- C
- C++
- Java
- PHP
- SQL
Backend
- FastAPI
- Node.js
- Express
- Prisma
- REST APIs
Frontend
- Next.js
- React
- Tailwind CSS
- Zustand
- Framer Motion
Data & Storage
- PostgreSQL
- MySQL
- MongoDB
- ChromaDB
- Supabase
- pandas
- NumPy
Tooling & Ops
- Git
- GitHub Actions
- Docker
- PHPUnit
- Vercel
- Jupyter
Education
Sep 2023 — Dec 2026
In progress · graduating Dec 2026
UAC 754100
UOW course 766
BCompSci · 144 cp
Bachelor of Computer Science — Artificial Intelligence & Big Data
University of Wollongong · Faculty of Engineering & Information Sciences
One degree, two campuses: commenced at UOW Dubai in September 2023, then transferred to the main Wollongong campus in February 2025 to finish in Australia. Coursework spans big data mining, generative AI, software methodologies, cybersecurity, and a year-long capstone combining NLP with program analysis.
Contact
Hiring for 2027 graduate roles or internships in Sydney? Let's talk.