PPDAT GitHub ↗

Pengyi Data Agent Team · Data Infrastructure

Reliable data.
Research ready.

PDAT discovers, evaluates and engineers external data into standardized, versioned assets for PAAT and the wider Pengyi Quant system.

Meet the teamInspect the pipeline ↓
Data Collector Agent robot
01 · Collect
Data Analyzer Agent robot
02 · Analyze
Data Engineer Agent robot
03 · Engineer
3

Production agents

1

Human audit gate

4

Data product layers

0

Silent repairs allowed

01 · The Agent Team

Three roles.
One governed flow.

Each Agent owns a distinct decision boundary. Discovery, evaluation and engineering remain separate so every dataset can be traced and challenged.

01
Data Collector Agent

Source intelligence

Data Collector

Translates a concrete data request into a candidate pool with coverage, access, cost, licence and stability evidence.

  • Source discovery
  • Access mapping
  • Candidate registry
02
Data Analyzer Agent

Independent evaluation

Data Analyzer

Scores candidate datasets for quality, timeliness, compliance, engineering cost and relevance to current PAAT research.

  • Quality scoring
  • Duplicate control
  • Primary / fallback decision
03
Data Engineer Agent

Production engineering

Data Engineer

Ingests only approved sources, validates the payload and publishes immutable, queryable data products with operational evidence.

  • Ingestion and normalization
  • Quality gates
  • Incremental serving

02 · Data Lifecycle

External data becomes
a governed asset.

INPUTData RequestMarket, fields, history, budget, use case
→
DISCOVERCollectorCandidate sources and access evidence
→
DECIDEAnalyzerScore, deduplicate, recommend
→
GATEHuman AuditFast, Review or Strict mode
→
BUILDEngineerIngest, validate, store and serve
01

Raw

Immutable vendor-form source records with ingestion lineage.

pd_raw
02

Clean

Normalized identifiers, calendars, fields and quality status.

pd_clean
03

Feature

Point-in-time, strategy-ready fields with availability rules.

pd_feature
04

Serving

Version-pinned APIs and exports consumed by PAAT.

Data API

03 · Human Governance

Automation moves fast.
Authority stays explicit.

Pengyi defines the request and owns high-risk approvals. Little G Cat watches the control plane: versions, quality failures, permissions and the next action.

Paid source → ReviewRestricted licence → ReviewLive dependency → Strict
Pengyi, human approval owner
PengyiHuman Decision Gate
Little G Cat, PDAT orchestrator companion
Little G CatControl Plane Watch

The Quant Production Chain

PDAT publishes trusted data.
PAAT turns it into research.

PDATDatasetSnapshot→PAATAlphaArtifact→PCPT
View repository ↗