03
03 Product Screens
Screenshots from live systems
A selection of screens from products and systems I have delivered. Select any image to view it in full.
03
Project Manager & Business Consultant / 2022 – 2026
The Selected version highlights representative projects for a quick read. The Full version includes additional projects and more implementation detail.
The Selected version highlights representative projects for a quick read. Each group still shows its full scale and links to the complete project list.
A selection of screens from products and systems I have delivered. Select any image to view it in full.
Open any project that interests you. The portfolio will switch to the Full version while keeping the current portfolio mode.
Agents that assist customers through chat. They use product knowledge, understand promotions, record orders, and hand over to a person when a question falls outside their scope.
The two totals combine verified counters from five systems: AB Beauty World, Lekofe Coffee, 1ManBiz, ABTravel, and Ladalat.
Beauty retail chain with 5 stores in Ho Chi Minh City and more than 4,000 SKUs.
A 12-store chain with several overlapping promotion types.
A marketing-tools platform with a catalog spanning several product categories.
A recipe-led sales flow: suggest a dish first, then sell the ingredients.
A travel agent that answered questions and recommended tours through chat.
Figures were verified against the project's operations summary.
A retail product advisory agent for chat channels.
Seven brands running on one agent system.
Shipping service advice and order tracking.
A preschool admissions agent that collected parent information and scheduled school visits.
A health-supplement product advisor for chat channels.
Replacing repetitive manual work with controlled workflows. Every step records its status, exceptions are visible in a dashboard, and failed operations can be rolled back.
The invoice total combines three facilities in one hospital system. Time saved comes from the shipment-update project and was reported by the operations team.
The former software provider stopped operating, while three facilities still needed to issue invoices every day.
The same work, shown under two operating models.
| Step | Old process | After automation |
|---|---|---|
| Data consolidation | Staff collected figures from several screens and copied them into a spreadsheet. | The BOT read data on schedule with no manual copying. |
| Invoice creation | Staff entered each invoice in MISA, with one person responsible for each facility. | The BOT worked in MISA following the specified business sequence. |
| Error checks | Errors were found after issuance and required corrective work. | Pre-issuance checks stopped errors and recorded the reason for each stop. |
| Status tracking | Staff had to ask the person responsible for each facility. | A shared dashboard showed real-time status. |
| Notifications | Sent manually when someone remembered. | Automatic email based on processing status. |
| Volume | Limited by available staff | 16,853 invoices across 3 facilities |
This comparison was reconstructed from the process description and operating figures in the project documents.
Updated Amazon shipment details in two internal systems at the same time.
Five SaaS platforms built to serve multiple clients through user accounts or service plans, with operating metrics tracked for each product.
Agent and user totals come from Conductify. The transaction-user figure comes from MentorX. Each metric is shown with the product it belongs to.
Turns an authenticated browser session into an agent that works directly on websites.
Community, courses, and mentor bookings on one platform, with Vietnamese payment support.
A low-code platform where users can build their own AI agents.
A desktop application that brings multiple AI providers and chat channels into one place.
This page shows the main workflow in the product interface.
Built a knowledge graph from enterprise documents, then answered questions using that graph.
This page shows the knowledge structure and the main workflow in the product interface.
Two types of work for related audiences: answering admissions questions for learners and digitizing paper records to reduce manual entry.
Program and form totals combine two projects at HIU.
More than 40 academic programs, 5 admission methods, and one peak season.
A Chrome extension that digitized admissions records and replaced manual entry.
An admissions agent for applicants and parents.
An admissions agent for a university system with several member institutions.
A learner-support agent for an online learning platform.
An information agent for a public-sector organization.
An online Q&A agent for citizens.
Current-state discovery, multi-level stakeholder interviews, solution design, and pricing for large clients—the work before the first line of code.
Clients in this group are anonymized by agreement; only their industry and scale are shown. The three figures above come from the same proposal for the diversified group on page E-02.
A technical research agent running inside the client's infrastructure, with no data leaving its servers.
| Package | Scope | Timeline | Investment |
|---|---|---|---|
| A — Foundation | AI infrastructure, a working first version, and baseline measurement. | About 6 weeks | ≈ USD 12,000 |
| B — Production rollout | Production-ready technical-bulletin extraction plus the Q&A collector. | About 4–5 months | ≈ USD 23,000–25,000 |
| C — Self-improvement | Both systems in production, a self-improvement loop, and client ownership. | About 8–9 months | ≈ USD 38,000–42,000 |
Content is shown within the agreed scope; the client's identity is withheld under a confidentiality agreement.
An enterprise-wide AI roadmap for internal infrastructure.
| Capability | Business problem | Acceptance metric |
|---|---|---|
| Internal research assistant | An internal chatbot searched the document repository and let users rate each answer. | Immediate answers across thousands of files |
| Price tracking | Collect, compare, and retain prices from several sources. | ≥90% accuracy, one run per day |
| Candidate sourcing | Scan recruitment channels and score candidate fit. | Query <5 seconds, keyword match ≥80% |
| Organization charts | Turn text into charts matching the group template. | Render <30 seconds, up to 500 nodes |
| Meeting assistant | Turn meeting audio into minutes and action items. | Word error rate ≤15%, action-item capture ≥80% |
| Supplier sourcing | Collect supplier data from public sources. | Successful extraction ≥85% |
Content is shown within the agreed scope; the client's identity is withheld under a confidentiality agreement.
High-volume, year-round recruitment across several provinces, with frequent employee turnover.
Impact figures are estimates from the proposal; the client's identity is withheld under a confidentiality agreement.
An AI transformation program spanning communications, factory operations, and internal training.
The 80% automation target comes from the solution proposal; the client's identity is withheld under a confidentiality agreement.
Payroll automation for a multi-branch chain, organized into four modules.
Impact figures are estimates from the proposal; the client's identity is withheld under a confidentiality agreement.
The work before AI: requirements, process mapping, ERP delivery, cross-border storefront operations, and sourcing.
The module total comes from the internal ERP project on page F-01. The impression increase comes from the three Alibaba storefronts on page F-04, rising from 91,000 to 390,000.
About 13 modules, from requirements through production launch and user training.
Product research and solution options for a North–South rail container operator.
Three agricultural export storefronts: macadamia nuts, cashews, and coconuts.
Sourcing and supplier or contract-manufacturer coordination across several product categories.
A custom toolkit for BA and PM work: requirements discovery, planning, modeling, specification, change tracking, and output quality control.
Counts were taken directly from my workstation on August 24, 2026. The toolkit supports the full BA and PM lifecycle, from discovery and planning to change management and quality control.
Each custom skill handles one specific BA task with a defined input and output. The six groups below cover the work from requirements discovery through document approval.
These 69 BA skills sit within a local library of 195 skills; the remainder cover technical work outside BA. Command names are retained as implementation evidence. The table shows representative commands rather than the full list.
Eight connected steps, with each output feeding the next. Before delivery, documents are reviewed against principles tied directly to implementation and handover.
Step 08 reruns checks across the full chain to find conflicts between documents; it does not create a new document.
These four principles apply to outputs from all eight steps, not to one specific project.
A custom automation layer handles repeated work: retaining project context, controlling risky actions, assigning work by role, and wrapping up each session.
| Group | Count | Purpose |
|---|---|---|
| Automated hooks | 34 | Thirty-four hooks run at 9 points in the session lifecycle, from loading context to wrap-up. They retain project rules, check risky write areas, and update work status without relying on memory. |
| Role-based agents | 18 | Eighteen roles have separate scopes and permissions, so each agent receives only the right work: planning, research, coding, testing, review, documentation, project management, interface design, or data administration. |
| Commands | 50+ | More than 50 commands package recurring work into reusable workflows: planning, coding, testing, review, documentation, presentation, context retention, and version management. |
| Output detail levels | 6 | Six levels tailor the same material to its reader, from business users who need a clear conclusion to technical users who need implementation evidence. |
The command count is shown as 50+ because the command set changes weekly. The counts of 34 hooks, 18 agents, and 6 detail levels were taken directly from the workstation on August 24, 2026. Screenshots are omitted from this group to avoid exposing client-project content.
Assistant workflows I built for personal use and to teach learners how to run them on their own computers: reading files, entering data into systems, running on schedule, and supporting role-specific work.
The library includes 20 screen-recorded scenarios and 56 scenarios across nine roles. Some videos and scenarios are simulations, and all displayed data is sample data.
Four assistant workflows I built for personal use and to teach learners how to run them on their own computers.
Five scenarios that turn scattered documents into usable decision-making data.
Images come from practical and simulated videos; all displayed data is sample data.
Eleven scenarios covering market research, clean customer data, content production, and sales tools.
Some scenarios are simulations; images use sample data.
Each scenario explains why a standard chatbot cannot complete the work—the line between an assistant that answers and one that takes action.
| Role | Scenarios | Representative example | Why a standard chatbot falls short |
|---|---|---|---|
| CEO | 8 | Track decisions and commitments after meetings | Requires long-term memory, structured storage, and automatic reminders. |
| Marketing Director | 8 | Analyze cross-channel marketing performance | Requires data from several sources to be combined and normalized. |
| Sales Director | 8 | Flag sales opportunities at risk | Requires continuous CRM analysis against historical patterns. |
| Sales Representative | 8 | Update the CRM after every interaction | Requires event capture, structured extraction, and write-back to the system. |
| Marketing | 8 | Produce a campaign landing page | Requires both copywriting and a working page. |
| Customer Support | 4 | Classify and route tickets automatically | Requires ticket-system integration, routing rules, and automated action. |
| Human Resources | 4 | Onboard new hires by role | Requires internal documents and a repeatable workflow for each person. |
| Finance | 4 | Reconcile invoices, contracts, and payments | Requires document extraction, matching, and exception handling. |
| Operations | 4 | Track process bottlenecks | Requires event data, process state, and early warnings. |
The full list of 56 scenarios appears only on screen. The print version keeps the nine-role summary so it fits on one page.
Content generated under brand control, video built from a written process, and a branded storefront run on Amazon.
The three control layers come from page I-01. The two production routes and the 15-minute mark are my own tests, timed on my own machine, recorded on page I-02. The three shorts are client work, made to a normal brief and schedule.
The brand guideline stops being a PDF and starts controlling what gets made.
An in-house product built to present to a bank. Vietnamese interface, simulated data, client not named.
I run social media content for a client brand: scripts, edits and the production process behind them. No dragging a timeline; each step is a written command that can be re-run.
| Route | Input | What the machine did | Time |
|---|---|---|---|
| From raw footage | One uncut 4K clip, 32 seconds | Speech to text, three takes kept, slow zoom added, audio cleaned, a headline banner made, captions with keywords highlighted, sound levelled | ≈ 15 min |
| From a written script | An eight-part script, no footage, no voice | Voice generated, footage found for each scene, royalty-free music picked, word-by-word captions built, exported vertical | ≈ 20 min |
Both routes have a step-by-step write-up, errors included.
Store and supply-chain operations, from product listings to inventory, costs, and profit and loss.
Choose based on how clear the requirement is, not on company size.
| Engagement | Best when | Work covered | Delivered outcome |
|---|---|---|---|
| Discovery and proposal | The problem is visible, but the right starting point is not clear. | Interview stakeholders at several levels, establish a baseline, map the current process, and prioritize the problems. | Current-state report, prioritized problem list, and an itemized proposal with costs. |
| Deliver one complete solution | The requirement is clear and a working system is needed. | Specify, build, test against business scenarios, train users, and launch in production. | A system running on real data, operating documents, and trained users. |
| Phased partnership | Several modules are needed and should be phased to control risk and cash flow. | Split delivery into separately accepted phases. A later phase starts only after the previous one passes. | Each module is handed over independently, with acceptance metrics written into the contract. |
Costs are based on the scope of each problem. Group E includes examples from real proposals.
Six stages, each with a verifiable output. The next stage does not begin until the previous one passes.
The baseline measured in step 01 is compared with the result in step 06. Without a baseline, there is no basis for claiming improvement.
| Period | What the client provides | Output |
|---|---|---|
| 01 · Discovery | Frontline users join interviews and demonstrate the real work. | A current-state report with baseline measurements. |
| 02 · Proposal | Agree on scope and priorities. | An itemized proposal with cost and timeline. |
| 03 · Specification | Approve the process, business rules, and acceptance metrics. | Requirements and acceptance metrics written into the contract. |
| 04 · Pilot | One real user group and real data. | A working narrow-scope version with error and exception logs. |
| 05 · Expansion | Operations staff join training. | A full-scope running system with user guides. |
| 06 · Acceptance | Review new measurements against the baseline. | Acceptance record against each agreed metric. |
This phase structure comes from the proposal on page E-02, where a large project was split into smaller, separately accepted parts instead of one all-in contract.
Terms made clear before work begins, so neither side has to guess when an issue arises.
These principles were applied in group E projects where contracts required on-premise model execution and commitment wording was reviewed before proposals were sent.
Four years of experience, including three years in project management and business analysis for AI products.
| Period | Role | Main work |
|---|---|---|
| 07/2025 – 05/2026 | Project Manager | Managed up to 6 AI projects at once, from presales and prototypes to live operations. Led presales and solution design for the two largest enterprise proposals in this portfolio. |
| 10/2024 – 06/2025 | Delivery Lead | Delivered more than 20 systems serving over 100,000 conversations. Reduced implementation time for one system from 2 months to 1 week. |
| 02/2024 – 09/2024 | AI Trainer / AI Engineer | Trained and refined AI agents, and built business workflows and knowledge bases for the first projects. |
| 06/2023 – 02/2024 | Team Lead | Led a research and development team delivering an internal ERP and a cross-border shipment-tracking workflow. |
| 09/2022 – 06/2023 | R&D Specialist | Delivered about 13 ERP modules from requirements through production launch and training. Researched transport-management products in Vietnam. |
| 07/2022 – 09/2022 | R&D Intern | Operated cross-border e-commerce storefronts, sourced products, and worked with suppliers. |
Enterprise-consulting clients are anonymized under confidentiality agreements.
A university foundation in quality management, followed by certifications in business processes and ERP.
Recognition received during university.
| Recorded | Awarded by | Content |
|---|---|---|
| Winner | EPIC · Event Planner | Planned and developed an event concept, sourced venues, balanced the budget, and pitched the proposal to a judging panel. |
| Second Prize | Faculty-level Student Research Competition | Student research project. |
| Distinction Scholarship | University of Economics Ho Chi Minh City | Scholarship for outstanding performance in Academic and Extracurricular Activities. |
| Top 5 in the major | University of Economics Ho Chi Minh City | Ranked across the major and graduated with distinction. |
This portfolio is updated regularly. Get in touch if you need the latest version or would like to discuss a specific project in more detail.