From conversations to intelligent agents.
I design and develop AI experiences that
feel natural to use.
Hi, I’m Trung — AI developer.
Projects · Stack · Contact · clean interface · product feel
From AI ideas to human experiences.
Conversational AI, agent workflows, and thoughtful product experiences.
The stack behind the intelligence.
The engineering foundation behind my AI applications — from model integration and backend logic to the interfaces people interact with.
Curious by nature.
Thoughtful by design.
I’m Nguyen Hoang Trung, a developer focused on applied AI. I explore how language models, intelligent agents, and thoughtful interaction design can work together to create better digital experiences. My focus is on the full experience: understanding intent, shaping the response, and making AI feel clear and natural to use.
From conversation to intelligent experiences.
Four explorations in conversational AI, agent behavior, prompt routing, and visual review. Each case study identifies what is implemented and what is still a concept.
TrungAI
AI application · API-connected chat
Conversational AI, with a clear path from question to response.
- Challenge
- Make a portfolio easier to explore through conversation instead of asking visitors to search every section.
- Approach
- A chat interface combines suggested questions, conversation history, and clear loading and error states.
- Implementation
- HTML, CSS, and JavaScript on the frontend; a Cloudflare Worker connects to OpenRouter and validates requests.
- Current outcome
- An API-connected chat experience is embedded here. Explore the interface below or visit the standalone website.
AI Simulation Lab
Interactive concept · Simulated agent behavior
Make an agent’s decisions easier to follow.
- Challenge
- Agent behavior can be difficult to understand when the intermediate steps are hidden.
- Approach
- A scenario interface places the request, decision timeline, and execution-style log in one view.
- Implementation
- Frontend interaction design in HTML, CSS, and JavaScript, using scripted scenarios and sample confidence values.
- Current outcome
- An interactive concept for reviewing agent behavior. The preview simulates decisions; it does not execute a live agent.
Scenario passed core checks. Manual review recommended before live execution.
> input parsed
> proxy_check: pass
> device_state: ready
> final_plan: generated
Prompt OS
Interactive concept · Prompt routing
Give an ambiguous request a clearer next step.
- Challenge
- Broad prompts often leave intent, task order, and expected output unclear.
- Approach
- An intent map, agent preset, and review panel illustrate how a request could become a structured plan.
- Implementation
- HTML, CSS, and JavaScript prototype with scripted routing states and a sample planning flow.
- Current outcome
- A working interface prototype of the routing experience. Classification and plans in this demo are illustrative.
Vague request converted into a 6-step automation brief with clear inputs, risks, and review points.
> vague_text: compressed
> actions: grouped
> risk_note: attached
> draft_status: review
Vision QA Agent
Interactive concept · Visual AI review
Turn visual feedback into an actionable review.
- Challenge
- Layout feedback is hard to act on when the issue, location, and severity are disconnected.
- Approach
- A visual review panel groups sample findings by spacing, contrast, and mobile layout, with a patch queue.
- Implementation
- HTML, CSS, and JavaScript prototype demonstrating the review interface with predefined findings.
- Current outcome
- An interactive visual QA concept. Findings are sample data; live screenshot analysis is not connected.
3 layout tasks
Fixes are grouped by visual impact instead of score percentage.
Patch readyThe engineering behind my perspective.
Additional work in device management and workflow interfaces, supporting my focus on AI experiences.
Android Device Manager
Engineering project · Device management
Supporting engineering work in device workflows.
- Scope
- A Windows application interface for Android device management, backup, restore, and repeatable device workflows.
- Technical context
- ADB and TWRP workflows are represented in the project. The embedded frontend demonstrates controls and status feedback.
- Preview status
- The browser preview uses sample device data; its buttons do not control a connected phone.
[12:13] Change info ready
[12:14] Proxy manager loaded
Automation Dashboard
Interactive concept · Workflow dashboard
A clearer view of repeatable work.
- Scope
- A dashboard concept for observing job queues, device status, and task history in one place.
- Implementation
- HTML, CSS, and JavaScript interface with interactive controls, status cards, and a sample activity log.
- Preview status
- The counters and success rate are demonstration data, not measured production results.
ADB target selected: device-08
Proxy check passed in 184ms
Random info profile generated
Task exported to history.csv
Good things
start with
a hello.
Building an AI product? Let’s explore your idea — from a conversational assistant to an agent-powered experience — and give it a thoughtful, human-centered interface.
Email [email protected] ↗My Favorite Side Quests
League of Legends
I started playing LMHT from Season 4. Up to now, that is about 12 years with the game.
PUBG
More than 5,000 hours played. A long-time game for aim, patience, focus, and survival decisions.
One Call Away
A song I like for its simple, warm feeling. Easy to replay while relaxing or working.
One Piece
Adventure, friendship, loyalty, dreams, and characters that keep moving forward.
A Record of a Mortal’s Journey to Immortality
A cultivation story with patience, strategy, progress, and the feeling of slowly becoming stronger.
Renegade Immortal
A darker cultivation vibe with persistence, conflict, and a long path of transformation.