Portfolio · 2026

Pengcheng Xu, builds intelligent systems.

I design and ship AI-powered applications and real-world products — where systems thinking, mobile craft, and language models meet.

Software Engineer/ AI Tools/ iOS/ Full-Stack/ Backend
Based
Irvine, CA
Education
MSWE · UC Irvine
Status
● Open to roles · 2026
Focus
Full-Stack · iOS · AI Engineering
scroll
02 · What I Build

Three things I build — with receipts.

Each of the next four featured projects lives in one of these buckets. The numbers are what shipped, not what was promised.

01

Hybrid retrieval, citation-grounded LLMs, multi-dimensional scoring. Designed for recall, latency, and refusal — not demo-day magic.

AI systems where retrieval beats fine-tuning

DevPulse · +23% Recall@5 · 50K docs
DeepNews-LLM · 90%+ accuracy · 1K+ articles/batch
02

On-device vision, AR navigation, real-time guidance. Apple's frameworks pushed past the obvious — interface as instinct.

iOS where intelligence feels native

Focuspace · −35% framing · SSC '26
MemoTree · 98.5% precision · ARKit
03

Smart contracts, async ingestion, cache strategy, reliability budgets. Boring infrastructure — measured, monitored, done well.

Backend that holds up under traffic

GreenTraceChain · 500+ TPS · 5 contracts
Yiwu Peisha (intern) · 10K+ users · −40% latency
03 · How I Think

A simple loop. Repeated obsessively.

Same three steps every time. Below each, a moment from a real project where the step paid off.

1

Understand the problem

Deeply. Before code, before architecture — what are we actually trying to make easier? Who's on the other side of the screen?

↪ DevPulse · week 1 Read 200+ engineering tickets and Slack threads to figure out what "searching code" actually meant to engineers — only then touched the indexer. Saved a month of building the wrong thing.
2

Design scalable & intelligent solutions

Pick the right primitives — embeddings, queues, caches, agents. Sketch the system before the implementation.

↪ DeepNews-LLM · scoring layer Chose AHP × entropy weighting over simple averaging — every dimension's contribution becomes transparent and reproducible. Reviewers can trace why a score is 42 vs 80, not just see the number.
3

Optimize performance & UX

Measure what matters: latency, recall, retention. Then make it feel inevitable to the person using it.

↪ Yiwu Peisha · Material Inquiry path Cache strategy rewrite + collapsed N+1 queries on Java/Spring Boot + MySQL → −40% latency. Held through the full-traffic ramp to 10K+ users without incident.
Featured · Project 01 AI · RAG Platform 2024 — 2025

DevPulse — making 50,000 documents
feel like one searchable mind.

— The Problem

Engineering teams write more than they can read. Important context lives buried in docs, postmortems, code comments, Slack threads. Search is keyword-shallow; LLMs hallucinate without grounding.

— What I Built

A retrieval-augmented platform that combines dense embeddings, hybrid keyword ranking, and a small reasoning layer — so questions return the source, not a guess.

+23%
Recall@5 lift over pure-dense
baseline · 50K-doc corpus
2 services
Spring Boot API + FastAPI worker
scaling independently
Retrieval dense embeddings (pgvector) + sparse BM25, fused via Reciprocal Rank Fusion — +23% Recall@5 on a 50K-document corpus
Architecture distributed multi-service: Java Spring Boot API layer + Python FastAPI AI worker — request handling decoupled from LLM inference, each scales independently
Streaming Redis Pub/Sub bridges backend inference and frontend, low-latency token delivery
Generation grounded prompts with citation enforcement; refusal on low-confidence
Workflow AI-assisted dev (Claude Code · multi-agent · design-to-code) for consistency across services
Stack Java · Spring Boot · Python · FastAPI · pgvector · PostgreSQL · Redis · Kafka
// Key wins:
  • +23% Recall@5 vs. pure-dense baseline
  • Multi-service architecture lets the AI worker scale without touching the API tier
  • Redis Pub/Sub eliminates polling, keeps streaming responses snappy
devpulse.io / search
auth/middleware/token_refresh.py
0.94
platform-services · main · last edit 4d ago
…on 401, the middleware retries with a refreshed token using exponential backoff (max 3 attempts). If refresh fails, returns…
docs/architecture/session-lifecycle.md
0.81
handbook · v3.2
token expiration is detected at the gateway layer; retry semantics differ between user and service…
incidents/2024-11-auth-storm.md
0.76
postmortems · resolved
…retry storm caused by misconfigured backoff on expired tokens; mitigation rolled out…
50,124 docs indexed ● synthesized in 412ms
— Why this stack

Hybrid retrieval (dense + sparse via RRF) lifted Recall@5 by 23% on a 50K-document corpus. Java Spring Boot for the API tier, Python FastAPI for the AI worker — each scales without touching the other. Redis Pub/Sub bridges them at streaming latencies.

+23% Recall@5 Spring Boot + FastAPI RRF fusion Redis Pub/Sub
Featured · Project 02 Blockchain · Carbon Trading 2024 · National Award Winner

GreenTraceChain — turning every gram of CO₂ into auditable, on-chain truth.

— The Problem

Carbon accounting and offset markets are riddled with opaque data: emissions are self-reported, ledgers can be quietly amended, and regulators see numbers days — sometimes months — after they matter. The "double-carbon" goal needs trust the current system can't provide.

— What I Built

A four-role carbon platform on ChainMaker (长安链): enterprises file emissions, auditors review & sign, regulators watch the ledger live, and admins mint carbon credits — all stitched together by five smart contracts that make every record immutable, traceable, and quantum-safe.

+25%
Network throughput
500+ TPS sustained
−35%
Avg query latency
via refined block indexing
Chain ChainMaker (长安链) permissioned consortium; per-org certificate identity; 500+ TPS sustained, +25% throughput vs. baseline
Contracts CarbonCoin · CarbonEmission · CarbonModel · CarbonTrade · Register (5 contracts)
Indexing refined block indexing + tight data structures cut average query latency by 35%
Backend Go for high-throughput tx ingestion; Java/Spring for the role-based services
Security ECDSA signatures, SHA-256 hashing, AES at rest — tamper-proof transaction trail
Frontend 4 separate web consoles (enterprise / auditor / regulator / admin)
// What the system gets right:
  • Led a 5-member team architecting the chain + service layer
  • Industry-specific carbon models (2 verticals shipped) for accurate accounting
  • CarbonCoin rewards convert verified reductions into tradable credits
  • Credit-scoring + clearing keeps the market honest under partial trust
  • Won First Prize at the 2024 China Service Outsourcing Innovation Competition (Aug 2024 · A21 · 云象网络)
chainmaker · greentrace explorer ● synced · block #2,358,401
Carbon offsets
14,820tCO₂
Verified by
4parties
Contracts
5active
block #2358399
0x3e2d…7c80
CarbonCoin · +480 ⊙
✓ minted
block #2358400
0x9c1e…4f0a
CarbonEmission · 1,420 tCO₂
✓ on-chain
block #2358401
0x7af3…b912
CarbonTrade · 200 tCO₂
✓ verified
contracts › CarbonCoin CarbonEmission CarbonModel CarbonTrade Register
2 industry models · post-quantum signed ● block sealed 1.4s ago
— Recognition · Leadership

Led a 5-member team architecting the chain + service layer. Won First Prize at the 2024 China Service Outsourcing Innovation Competition (题号 A21 · 云象网络) with team 你的外包我来包.

First Prize · Aug 2024 5-member team lead 500+ TPS · +25% ECDSA · SHA-256 · AES
Featured · Project 03 AI · LLM Verification 2025

DeepNews-LLM — scoring news on eight dimensions
before you trust it.

analyzing · live example-news.com · 2.4k words · zh-CN
"Scientists prove a single cup of tea reverses aging in seven days"
internal · 5 dimensions
Title relevance8.0
Logical consistency7.0
Factual accuracy3.2
Subjective tone3.8
Causal reasoning5.5
external · 3 dimensions
Source credibility4.8
Fact-check status2.5
Cross-source verify4.2
42/100
low credibility · 3 critical issues AHP × entropy weighted · streamed in 1.8s
— Sample dimension prompt
// internal · 03 · factual_accuracy
retrieve(article, k=8) → evidence
extract(article)         → claims[]
score = max(0, 10 − 0.5 × |conflicts|)
refuse if confidence < 0.6
— The Problem

Generative AI made fabricating a believable headline a one-prompt task. Single-signal fact-checkers — source blacklists, keyword filters — miss the patterns that make a story feel true while being false. Readers need a second opinion that thinks, not just blocks.

— What I Built

A LangChain-based agent that scrapes the article, then scores it on five internal dimensions (relevance, logic, facts, tone, causality) plus three external ones (source, prior debunks, cross-source corroboration). AHP × entropy-weighted aggregation gives the final verdict — streamed back over SSE as the agent thinks.

90%+
Classification accuracy
over labeled benchmark
−40%
End-to-end latency
via async streaming
Backend FastAPI + Uvicorn (async ASGI), Server-Sent Events streaming — −40% latency
Agent LangChain AgentExecutor with Pydantic models + dynamic tool routing — −35% model-call overhead
Scoring 8-dimension framework (5 internal + 3 external) weighted via AHP × entropy for objective, reproducible verdicts — 90%+ accuracy
Crawling Selenium queue with rate limiting + boilerplate removal — 1K+ articles/batch · <5% variance across re-runs
LLMs OpenAI primary; SiliconFlow fallback for domestic-network resilience
Search Google Search API + DuckDuckGo for cross-source verification
Storage MongoDB (Motor async) for history; Redis for hot-path cache
// Modes the API exposes:
  • POST /analyzeWeb · full 8-dimension report from a URL
  • POST /chat · conversational follow-ups with persisted session history
  • POST /chat/resultQuestion · ask why a specific dimension scored what it did
  • Microservice-shaped: gateway · agent · scoring · extraction split for independent scaling
Featured · Project 04 iOS · Computer Vision 2025

Focuspace — a camera that sees with you,
not just for you.

Focuspace iPad camera — sunset boathouse scene with breathing-dot composition guidance, 9 template buttons, and side toolbar
— Design philosophy
Move until it feels right.

No language, no menus — just a breathing dot that tells you where to point. The simpler the cue, the more the photo feels like yours.

Swift Student Challenge 2026 9 templates
— The Problem

People who travel with their iPad or phone don't take bad photos because their camera is bad — they take bad photos because nobody ever taught them composition. The viewfinder is silent. Symmetry, rule of thirds, leading lines… you either know them, or you don't.

— What I Built

An iPad camera app submitted to Swift Student Challenge 2026. Pick one of nine composition templates; a breathing white diamond floats around the viewfinder, telling you exactly which way to move. Move until it feels right — no jargon, no menus.

−35%
Avg shot adjustment time
vs. native camera UX
+28%
Subject-lock consistency
−40% overlay jitter under motion
Stack Swift · SwiftUI · AVFoundation (live capture) · Vision (per-frame analysis) · FoundationModels (on-device)
Templates 9 composition rules with overlay geometry: Symmetry · Center · Leading Lines · Spiral · Framing · Negative Space · Headroom · Triangle · Thirds
Tracking Vision-based pipeline integrating face analysis, object tracking, stabilization — +28% subject-lock consistency, −40% overlay jitter
Coach hybrid AI: rule-based scoring + on-device FoundationModels for scene understanding & template recommendation — +32% recommendation stability
Outcome overall shooting efficiency +22%; pure-algorithm fallback keeps the app working on older OS
Composition Lab per-template tutorial screen with reference photos & "Start Shooting" launcher
Capture rule overlay never written to the saved photo
// Design philosophy:
  • "Move until it feels right" — no language, just a dot you chase
  • Submitted to Apple's Swift Student Challenge 2026
  • Settings expose Preview Grid · Guidance UI (Scope/Missing/Arrow) · Template Overlay · Foundation Models toggle · Debug HUDs
  • Ships as a single Swift Playground (.swiftpm) — Xcode or iPad direct
09 · AI × Engineering

I don't just use build with AI
— I build systems with AI as a primitive.

A · RETRIEVAL

Production RAG & LLM pipelines

Embeddings, hybrid search, reranking, citation-grounded generation. Designing for recall, latency, and refusal — not demo-day magic.

B · AGENTS

Multi-agent coding workflows

Planner / executor / verifier loops that turn vague intent into reviewed pull requests. Tooling, sandboxing, observability.

C · DX

AI-assisted development

Heavy hands-on with Claude Code, Codex, and Cursor. I treat the model as a collaborator with a code review checklist.

10 · Experience

Real systems. Real users. Real numbers.

The work below shipped to production and was measured by people I had to keep happy.

Backend Developer Intern
Yiwu Peisha Network Technology Co.
Sept 2023 — Sept 2024 · Yiwu, China

Owned backend modules across the company's commodity-trading platform — Material Inquiry System, Product Library CRUD, and the price-trend visualization API. Took features from spec to production, then iterated on latency and reliability under real traffic.

// stack: Java · Spring Boot · MyBatis · MySQL · REST
  • 10K+
    Active users on the Material Inquiry System I shipped.Sustained through full-traffic ramp without incident.
  • −40%
    Query latency reduction on the hottest path.Cache strategy + query plan rework + N+1 elimination on Java/Spring Boot + MySQL.
  • +30%
    Reliability gain on the Product Library CRUD & REST surface.Idempotent writes, retry-safe APIs, defensive validation at the boundary.
  • −20%
    Latency cut on the price-trend visualization module & its API.Faster UI response — let traders make decisions in real time.
11 · Skills

Three lists, kept short on purpose.

I can hand you a spreadsheet of acronyms. I'd rather show you what I reach for first.

A · Languages

Things I think in.

  • Java · Swift · Python · Go
  • C++ · JavaScript / TypeScript
  • SQL
B · Frameworks

Things I build with.

  • Spring Boot · MyBatis · Gin · Gorm
  • SwiftUI · AVFoundation · ARKit
  • React · Vue · Node.js
  • FastAPI · LangChain
C · Data & AI

Where intelligence lives.

  • PostgreSQL · MySQL · MongoDB · Redis
  • Elasticsearch · pgvector
  • Core ML · Create ML · FoundationModels
  • PyTorch · TensorFlow · Hugging Face
D · Tooling

Boring infra, done well.

  • Linux · Git · Docker · Kubernetes
  • Kafka · REST · Nginx
  • Postman · Dialogflow
  • Claude Code · Codex
12 · Recognition

A few shiny things.

National prizes, regional firsts, an Apple certification, and a top-1% scholarship — receipts for the work above.

  • Aug 2024

    First Prize · National

    2024 China College Service Outsourcing Innovation & Entrepreneurship Competition — National First Prize for GreenTraceChain.

    National · Team lead
  • 2024

    National Innovation Project

    National College Student Innovation & Entrepreneurship Training Program — project approved at the national level under China's MOE.

    National · PI
  • Nov 2024

    National Scholarship

    China National Scholarship — awarded to the top 1% of students at the university level.

    Top 1% · University
  • 2025

    First Prize · East China

    2025 Mobile Application Innovation Competition — regional First Prize in the East China division.

    Regional · iOS
  • Jul 2023

    Second Prize · East China

    2023 Mobile Application Innovation Competition (East China division) — for an early iOS prototype that seeded MemoTree.

    Regional · iOS
  • 2023

    Third Prize · Zhejiang

    2023 Zhejiang Provincial College Service Outsourcing Innovation Application Competition — provincial Third Prize.

    Provincial · Team
  • Sept 2023

    Apple Certified

    App Development with Swift Associate — Apple's international certification for foundational iOS app development.

    International cert
13 · Let's talk

Let's build something
impactful together.