Nguyen Manh Tu
VIEN
Portrait of Nguyen Manh Tu

Software Engineer· AI Engineer

Hanoi, Vietnam

Building reliable backends and realtime AI experiences.

Contact
VCareer / View case study

150+ pilot learners

ABOUT / HOW I BUILD SYSTEMS

Foundation

From Backend. AI is the next step forward.

More than two years in Backend gave me a foundation in APIs, data, business logic, and how production systems behave. I bring that foundation into LLMs, RAG, recommendation systems, and AI Agents to build AI systems that work in practice.

Operating principle

Start with the real problem rather than the technology.

I first clarify the user's problem and what success looks like, then build the simplest solution that can test the idea. Architecture, performance, and AI complexity evolve only after real usage provides evidence.

How I make decisions

  1. Understand the problem
  2. Prove the simplest version
  3. Observe real usage
  4. Scale with evidence

What good engineering means

Good engineering is not the system with the most technology. It is simple enough to maintain and strong enough to solve the right problem.

04 / FLAGSHIP PROOF

Product live · further development pending university funding

AI Career Development Platform

VCareer

Connecting CV preparation, job criteria, and realtime AI interviews in one actionable practice workflow.

150+real learners in the VinUni pilot

What I directly owned

  1. LiveKit + WebRTC baseline
  2. CV analysis and CV-to-JD matching
  3. JD Builder

PRODUCT WORKFLOW / 06 SCREENS

From profile to feedback.

Product interface snapshots · values shown inside are not used as portfolio metrics.

  1. SCREEN 01 / 06

    PRODUCT CONTEXT

    Landing

    The entry point to VCareer's journey. Numbers shown in the interface are product demo data only.

    VCareer platform landing page

  2. SCREEN 02 / 06

    DIRECT SCOPE · ANALYSIS LAYER

    CV Builder

    The CV analysis and scoring layer is shown inside the builder.

    VCareer CV building and analysis interface

  3. SCREEN 03 / 06

    DIRECT SCOPE

    CV / job matching

    CV-to-JD comparison makes fit, gaps, and actionable recommendations visible.

    VCareer CV and job description matching interface

  4. SCREEN 04 / 06

    DIRECT SCOPE · BASELINE

    Live interview

    The LiveKit/WebRTC baseline carries the realtime AI interview session; the complete experience is team product context.

    VCareer live AI interview interface

  5. SCREEN 05 / 06

    PRODUCT CONTEXT

    Interview review

    The learner feedback loop after an interview, presented as context for the wider team product.

    VCareer post-interview review and feedback interface

  6. SCREEN 06 / 06

    PRODUCT CONTEXT

    Dashboard

    An overview of the learner journey and functional paths in the VCareer pilot system.

    VCareer learner dashboard

REALTIME PATH

BROWSERLIVEKIT / WEBRTC

05 / SYSTEMS & EVIDENCE

05 / SYSTEMS & EVIDENCE

AI RESEARCH SYSTEM

ScholarAI

AI-augmented literature discovery and reading workspace

SOLO PROJECT · END-TO-END ENGINEERING

From finding the right paper to tracing an answer back to source: retrieval, PDF reading, notes, citation-aware RAG, and evaluation share one workspace.

One codebase. The complete loop.

  1. Dense + BM25 retrieval fused through Qdrant RRF.

  2. Hierarchical RAG, SSE responses, and citations linked back to PDF pages.

  3. A LangSmith harness for QA, retrieval, faithfulness, and refusal.

ENGINEERING EVIDENCE / 03

FastAPI · Qdrant RRF · PostgreSQL · OpenAI · LangSmith

  1. SIGNAL 01 / 03

    RETRIEVE

    Search by meaning and exact terms.

    Dense embeddings and BM25 meet in Qdrant RRF, preserving semantic relevance alongside acronyms, author names, and exact-match queries.

    ScholarAI academic search interface with semantic retrieval and filters

  2. SIGNAL 02 / 03

    GROUND

    Answers return to source.

    Hierarchical RAG retrieves PDF context, streams the response, and attaches citations that reopen the relevant document location.

    ScholarAI PDF reader and citation-aware chat interface

  3. SIGNAL 03 / 03

    EVALUATE

    Measure instead of assume.

    Two LangSmith runs record QA and refusal behavior. Values remain visible as experiment output, not repackaged as an improvement claim.

    ScholarAI QA and retrieval benchmark results in LangSmith
    QA / RETRIEVALTracks answer presence, citation accuracy, correctness, faithfulness, retrieval hit rate, and latency.
    ScholarAI refusal guardrail benchmark results in LangSmith
    REFUSAL / GUARDRAILTests out-of-scope refusal behavior alongside latency and error rate for each run.

BACKEND FOUNDATION ARCHIVE

CAPSTONE · BACKEND LEAD · BUILT BEFORE AI-ASSISTED CODING

Financial Planning

Internal planning, expense approval, and financial reporting system

I led the backend for a role-aware financial workflow spanning authentication, authorization, planning terms, expenses, reports, and scheduled processing.

THREE DEPLOYMENT BOUNDARIES

  1. APISpring Boot · SQL ServerREST APIs, financial rules, and Redis-backed OTP, token, user-detail, and authority state.Open API source Opens in a new window
  2. WEBReact · ViteRole-aware interfaces for plans, approvals, master data, and reporting.Open Web source Opens in a new window
  3. WORKERSpring scheduled jobsSeparate recurring jobs for term state and annual-report aggregation.Open Worker source Opens in a new window

ARCHIVE INTERFACE / 02

Archived Financial Planning dashboard interface
DASHBOARDA snapshot of planning terms, departments, and demo expense allocation.
Archived Financial Planning report interface
FINANCIAL REPORTThe reporting surface connecting planned values to the approval workflow.

Source archive · No live demo or preserved runtime dataset

06 / EXPERIENCE & RECOGNITION

2019 → 2026

Career journey.

From an Information Technology foundation, through production backend work, to practical AI training.

DEVELOPMENT TRACE / 03 RECORDS

  1. 01 / FOUNDATION2019 — 2024

    Bachelor of Information Technology

    FPT University

    Hoa Lac, Hanoi

    Bachelor's program in Information Technology at the Hoa Lac campus in Hanoi.

  2. 02 / PRIMARY EXPERIENCEMAY 2024 — MAR 2026

    Software Engineer

    FPT Software

    Backend systems · 4–5 person sub-team

    Developed backend services for an HR system and helped coordinate a backend sub-team.

    WORK SCOPE

    1. Built APIs and business logic with Java, Spring Boot, and PostgreSQL.

    2. Implemented batch jobs and asynchronous processing with AWS SQS; used Docker and LocalStack for local development and testing.

    3. Coordinated work and reviewed code for a 4–5 person backend sub-team.

    SYSTEMS & TOOLSJava · Spring Boot · PostgreSQL · AWS SQS · Docker · LocalStack · GraphQL · Testing

  3. 03 / AI DEVELOPMENTAPR 2026 — JUL 2026

    Practical AI Talent Program — Foundation

    VinUniversity × Vingroup

    12 weeks · SFIA skills framework

    A hands-on program focused on AI thinking, AI ethics, AI assistants, and problem-solving under VinUni faculty.

RECOGNITION INDEX / 03

Recognised outcomes.

VinUniversity closing-ceremony stage showing Team VCareer as a featured project alongside the VCareer and WonderLens award results.
VINUNIVERSITY CLOSING CEREMONY

The VinUniversity closing-ceremony screen recognised Team VCareer and recorded both Codex Community Hackathon Hanoi 2026 results.

DOCUMENT REGISTER / MANUAL SELECT

01 / 03
01 / FLAGSHIP

VCareer

2nd Prize · Track 4: Transform with Codex

$5,000 in OpenAI API credits · Codex Community Hackathon — Hanoi

RELATED RECOGNITION / 02

  1. 02 / HACKATHONWonderLens

    1st Prize · Track 1: Market Scale

    Codex Community Hackathon — Hanoi

  2. 03 / VINUNIVERSITYVCareer

    Featured project

    Closing ceremony at VinUniversity

07 / CAPABILITIES & CONTACT

Capabilities anchored in evidence.

Each capability routes back to the project or role that demonstrates it.

  1. Backend systems

    APIs, business logic, relational data, asynchronous processing, and scheduled work.

    TECHNOLOGIES & METHODS

    • Java
    • Spring Boot
    • PostgreSQL
    • SQL Server
    • Redis
    • AWS SQS
    • Docker
    • LocalStack

    RELATED EVIDENCE

  2. Realtime AI experiences

    Realtime session foundations for a live AI interview experience.

    TECHNOLOGIES & METHODS

    • LiveKit
    • WebRTC

    RELATED EVIDENCE

  3. Retrieval & evaluation

    Hybrid retrieval, source-linked citations, and verifiable evaluation loops.

    TECHNOLOGIES & METHODS

    • FastAPI
    • Qdrant
    • BM25 / RRF
    • RAG
    • LangSmith
    • PostgreSQL

    RELATED EVIDENCE

  4. End-to-end product delivery

    Connecting product workflow, frontend, backend, and a usable delivery path.

    TECHNOLOGIES & METHODS

    • TypeScript
    • React
    • Next.js
    • FastAPI

    RELATED EVIDENCE

OPEN A CHANNEL

EVIDENCE CONVERGED → EMAIL

Have a system worth building?

I’m interested in reliable backends, realtime AI products, and problems that turn models into real experiences.

DIRECT CHANNEL / EMAIL

OTHER ROUTES

Portfolio / 2026

VIEN

NguyenManhTu
BackendRealtimeAI
Resolving critical scene

Backend · Realtime · AI

Nguyen Manh Tu — Systems in Focus