YN Software · AI

Open to new work

Software Engineer / AI Engineer

Engineering Software.
Building Intelligence.

I build for the web — frontend interfaces, full-stack services, and the systems that hold them together. Lately my work has moved toward applied AI: LLM-backed applications, retrieval pipelines over real documentation, and agents that do useful, bounded work instead of demos.

Explore Projects Get in Touch
Focus
Frontend · Full-stack · AI engineering
Working in
TypeScript · Python · Go
Based in
City, Country

01 / About

Engineering foundations,
applied to intelligent systems.

My background is ordinary software engineering, and I mean that as a compliment: readable code, clear boundaries, interfaces that behave under pressure, and architecture that someone else can pick up six months later without a guided tour.

On the frontend that means accessible, performant interfaces and component systems that stay consistent as a product grows. On the backend it means well-shaped APIs, honest error handling, and integrations that fail loudly rather than quietly.

The AI work grew out of the same instinct. An LLM is a component with unusual failure modes, not a magic layer — so I treat retrieval, evaluation, and guardrails as engineering problems. I build RAG pipelines where chunking, embeddings and ranking are deliberate choices, and agentic workflows that are scoped, observable, and reversible.

I'm most useful where product thinking, interface craft and systems work overlap — turning a vague "could AI do this?" into something shippable.

02 / Selected work

Projects where the hard part
was the engineering.

Featured / In development

Documentation RAG Assistant

Problem
Internal documentation is searchable but not answerable. People find a page, not the paragraph, and give up — or worse, ask a general-purpose chatbot that invents an answer.
Solution
A retrieval-augmented assistant that ingests documentation, chunks it with structure awareness, embeds it into a vector index, and reranks candidates before generation. Answers are assembled only from retrieved passages and always render their citations, so an unsupported claim is visible rather than plausible.
Stack
Python · FastAPI · embeddings · vector database · reranking · streaming UI

Repository — add link Live demo — add link

Prototype

Repository Analysis Agent

Problem Onboarding to an unfamiliar codebase, and preparing it to ship, is slow manual archaeology.

Solution An agent that walks a repository, infers its stack and entry points, and proposes the container and deployment configuration — each step a discrete tool call with an inspectable trace, so a human approves before anything is applied.

Python · tool-calling LLM · AST parsing · Docker · CI

Repository — add link

Exploration

Kubernetes Insight

Problem Cluster state is readable but not legible: the signal lives across dozens of resources and nobody reads it all.

Solution A read-only analyser that collects workload, resource and configuration data, flags misconfiguration patterns, and summarises each finding in plain language with the exact manifest it came from.

Go · Kubernetes API · Docker · LLM summarisation

Repository — add link

Academic research

Retrieval-Augmented Defect Prediction

Problem Defect-prediction models judge a code change in isolation, ignoring the history of changes that already look like it.

Solution A study of whether retrieving historically similar code and defect records as context improves a model's judgement about a new change — treating retrieval as a feature source rather than a chat convenience, and reporting the result honestly either way.

Python · embeddings · vector search · evaluation harness

Paper / write-up — add link

Professional work

Frontend Architecture Modernisation

Problem A long-lived frontend where every screen had re-invented its own buttons, spacing and states — slow to change and inconsistent to use.

Solution Consolidated the interface into a token-driven component library with documented states, keyboard and screen-reader behaviour, and incremental migration so feature work never stopped.

TypeScript · Vue · design tokens · CSS architecture · accessibility

Case study — add link

03 / Expertise

Tools I actually reach for.

01

Frontend

  • HTML5 & semantics
  • CSS3, Grid, container-aware layout
  • JavaScript (ES2022+)
  • TypeScript
  • Vue
  • Design systems & tokens
  • Accessibility (WCAG AA)
  • Performance & Core Web Vitals

02

Backend

  • Python
  • FastAPI
  • Go
  • REST & streaming APIs
  • PostgreSQL
  • Background jobs & queues
  • Third-party integrations
  • Testing & API contracts

03

AI engineering

  • LLM application design
  • Prompt & context engineering
  • Embeddings & semantic search
  • RAG pipelines
  • Vector databases
  • Agent frameworks & tool use
  • Structured output & validation
  • Evaluation & guardrails

04

Infrastructure

  • Docker & containerisation
  • Kubernetes
  • Cloud platforms
  • CI/CD pipelines
  • Git & trunk-based workflow
  • Linux
  • Observability & logging
  • Infrastructure as code

No percentages, no five-star ratings — depth is a conversation, not a bar chart.

04 / Experience & research

A short, accurate history.

  1. 20XX — Present

    Role title at Employer name

    Production engineering

    Frontend and full-stack delivery: shipping product features, hardening interfaces, and shaping the shared component layer other teams build on.

    TypeScript · Vue · Python · Docker · CI/CD

  2. 20XX — 20XX

    Previous role at Employer name

    Production engineering

    Built and maintained web applications end to end — interface work, API integration, and the deployment path that carried them to users.

    JavaScript · REST APIs · PostgreSQL · Linux

  3. 20XX — Present

    Applied AI experiments — independent

    Personal projects

    Self-directed work on LLM applications: retrieval pipelines over real documentation, tool-using agents with bounded scope, and small evaluation harnesses to tell improvement from vibes.

    Python · FastAPI · embeddings · vector databases · agent frameworks

  4. 20XX — 20XX

    Degree / programme — Institution name

    Education & academic research

    Coursework in software engineering fundamentals, with research interest in retrieval-augmented approaches to software defect prediction.

    Software engineering · machine learning · information retrieval

Entries marked in dashed type are placeholders — replace them with verified roles, dates and institutions before publishing.

05 / Contact

Let's build something
that actually thinks.

Open to software engineering and AI work — product frontends, full-stack services, RAG systems and agent tooling. If you have a problem that looks like one of the projects above, or one that doesn't, say hello.

you@example.com

No contact form — mail reaches me faster and nothing here needs JavaScript.