Ygrowly

YUGUANG LIU · AI APPLICATION DEVELOPER · CLASS OF 2027

I bring AI into real workflows—and engineer the system around it.

Focused on agent engineering and Python backends: trustworthy data, controlled execution, provable results. I led runtime governance and the eval loop for EnergyOps Agent during my internship, and built RuleArena to turn rule-change verification into release gates — what code can guarantee deterministically stays out of the model.

Explore Selected Systems ↓ Read Latest Writing →
  • Class of 2027
  • AI Application / Agent Backend
  • Internship through graduation
  • Open to opportunities
Recent work
  • Kingsoft EnergyOps
  • DataSphere
  • RuleArena
  • PayTrace
ygrowly — terminal
$whoami
Yuguang Liu / YgrowlyAI Application Developer · Data Systems
$ls systems/
energyops-agentai-bi-platformrulearenapaytrace [lab]
$help
projects experience writing github contact
$type help, projects, or contact

01 SELECTED SYSTEMS

RUNNING · VERIFIED

Not feature collections, but systems that can be operated and verified.

01

Enterprise Agent System · 2026

EnergyOps Agent From campus meter readings to trusted ledgers and governed agent operations.

A campus energy operations platform: raw cumulative readings from hundreds of meters become three trustworthy ledgers — usage, anomalies, and settlement — with a permission- and confirmation-bound conversational agent inside WPS Comate.

Business problem: unstable cumulative readings must become auditable statistics and settlement, while the agent must operate business capabilities without overstepping. I led the design and delivery of agent runtime governance and the eval loop; my mentor owned business direction.

  1. Cumulative readings
  2. Quality states
  3. Interval usage
  4. Layered aggregation
  5. Diagnosis
  6. Alert loop
  7. Agent calls
Data Quality
Seven quality states plus a trusted whitelist; only trusted intervals enter aggregation — no unsupported estimates.
Runtime Governance
Dozens of MCP tools consolidated into six capability bundles with R0/R1/R2 risk tiers and Before/After Tool hook chains.
Recoverable Ops
Idempotency windows, persisted task state, and startup backfill — restarts and out-of-order data leave no aggregation gaps.
system / 01 Enterprise Agent System · 2026
Cumulative readings Quality states Interval usage Layered aggregation Diagnosis Alert loop Agent calls
Eval
replay eval · task completion ≈ +12pt
Security
all internal privilege cases blocked
Perf
core API P95 ≈ −86% (fixed query set)
Ops
manual backfills ≈ −70% · scheduler 99%+
Pipeline
raw → interval → hourly → daily

Python · FastAPI · PostgreSQL · MCP · WPS Comate · pytest

View system case →

02

Enterprise AI Data Platform · 2025–2026

DataSphere AI BI Natural-language analytics with sources, definitions, and verifiable results.

An AI data-analysis and collaboration platform for teams without dedicated data staff: ask in natural language, and results bind evidence and metric definitions before settling into tables, reports, and dashboards.

Business problem: a SQL query executing is not the same as the business answer being right — ambiguous metrics must be clarified and results must be reproducible. As a core developer on a small team, I owned the semantic layer and domain routing, query safety, and evidence binding, and supported the lightweight lakehouse.

  1. Intent routing
  2. Schema retrieval
  3. Semantic parse
  4. SQL generation
  5. Safety checks
  6. Evidence binding
  7. Charts & reports
Semantic Layer
Industry domain packs hold metric definitions and term aliases; a two-level intent contract separates business understanding from SQL implementation.
Query Safety
Layered defenses — permission context, governed tools, read-only SQL AST review, and output masking — with tenant isolation injected server-side.
Evidence Binding
Ambiguous metrics must be clarified; charts bind call_id + query_fingerprint + result_hash and degrade instead of blocking.
system / 02 Enterprise AI Data Platform · 2025–2026
Intent routing Schema retrieval Semantic parse SQL generation Safety checks Evidence binding Charts & reports
Eval
end-to-end success ≈ +13pt (cross-industry set)
Recall
schema recall ≈ +14pt (fixed set)
Safety
all adversarial requests blocked · ~3% false blocks
Cost
analysis context tokens ≈ −80%
Trust
charts bind call_id + fingerprint + hash

Python · FastAPI · PostgreSQL · DuckDB · Parquet · R2 · Univer · MCP

View system case →

03

Personal Project · Independent Build

RuleArena Turning rule-change verification into release gates.

An adversarial verification platform for e-commerce rule changes: natural-language rules compile into human-confirmed executable contracts, a governed agent searches for high-risk action sequences, a clean environment replays them over real HTTP, and a deterministic oracle adjudicates — distilling minimal, reproducible counterexamples.

Business problem: promotion, refund, points, and membership rules change faster than manual regression can follow. The agent finds paths humans did not think of; deterministic code proves whether a path is actually broken.

  1. Rule change
  2. RuleSpec contract
  3. Human confirm
  4. Adversarial search
  5. Sandbox replay
  6. Oracle verdict
  7. Minimal counterexample
  8. Release gate
Anti Self-Proof
Candidate risk ≠ confirmed bug — the agent never touches ground truth; only a sandbox replay plus an oracle violation confirms an issue.
Bounded Exploration
No dynamic expression evaluation; a fixed set of domain primitives only, and unconfirmed rules never enter attack runs.
Regression Assets
Minimal counterexamples bind a full evidence chain, export to pytest in one click, and repaired versions replay historical cases.
system / 03 Personal Project · Independent Build
Rule change RuleSpec contract Human confirm Adversarial search Sandbox replay Oracle verdict Minimal counterexample Release gate
Eval Set
design · 24 cases, 16 dev + 8 hidden (isolated)
Mechanism
measured · 0 false positives · 0 ground-truth leaks · 3/3 same-version replays
Gate
counterexample → pytest regression → release gate

FastAPI · PostgreSQL · Redis · Explicit FSM · Delta Debugging · pytest

View system case →

02 EXPERIENCE

2025 → NOW

Connecting data, backend systems, and AI inside real projects.

Kingsoft

AI Development Intern · EnergyOps Agent

Agent runtime governance, trusted data pipelines, and the eval loop

Shenzhen Huize Zhiyuan

AI Application Development Intern · DataSphere

Semantic layer, domain routing, query safety, and evidence binding

Chengdu Qidian Tuojie

Full-stack Intern · Ovanta

Content system, payment pipeline, and multi-site delivery

LAB & OPEN SOURCE

Live

Ovanta

Self-service platform for visas and overseas identity

 Open Source

PayTrace

Payment anomaly diagnosis agent · simulated data and fault injection

03 WRITING

UPDATING

Notes on how systems are designed, verified, and corrected.

Blog posts are being organized — browse Notes →

Notes

Notes are being organized — meanwhile, browse Blog →

04 · A SYSTEM ECHOES

PLAYBACK

System Echoes

Verifiable evidence left by real systems, not adjectives.

Cumulative readings pass seven quality states before entering aggregation — unconfirmable intervals keep their raw facts instead of unsupported estimates.

EnergyOps Agent · trusted data foundation

04 · B EXPLORE BEYOND

BROWSE

Explore Beyond This Page

Four directions to keep browsing.

01

Things I Built

Full cases and evidence for EnergyOps, DataSphere, RuleArena, and PayTrace.

 02

Writing Along the Way

How systems are designed, verified, and corrected — blog and notes.

 03

Experiments in Flight

Internships, the lab, and open-source practice over time.

 04

About Me

Education, stack, and how I work — plus ways to reach me.

05 PROFILE

PUBLIC RECORD

Education, stack, and how I work.

Education
University of South China · Data Science and Big Data · 2027
Focus
AI Application / Agent Backend (Python)
Backend
Python · FastAPI · PostgreSQL · Redis · DuckDB
Agent Engineering
MCP · Hook-chain governance · Eval · Trace · Observability

What code can guarantee deterministically stays out of the model.

More about me →

06 CONTACT

ONLINE · OPEN

Let’s talk systems, agents, and what comes next.

I’m looking for 2027 new-grad opportunities in AI application and agent backend engineering (Python), and I can start early and keep interning through graduation — always open to conversations about agents, data systems, and open source.