Technology
Not low-latency market trading — research is the architecture.
ZhiQing PreFounder is fundamentally different from securities-trading architecture: the core is retrieval-augmented generation, a structured numerical engine, evaluation sets and content-safety filters, plus a full audit trail.
Six-Layer Stack
Research → Retrieval → Agents → Evaluation → Delivery → Audit
Six-Layer Stack
Testable, auditable, and reproducible at every layer.
Data Layer
A structured industry database (public reports, financials, policy, patents, hiring trends) + isolated stores for users' private data
Retrieval
Hybrid vector search + BM25 + relationship graph, with a Re-rank Agent reordering evidence fragments
Agent Mesh
12 specialized agents (industry research / competitor / policy / finance / legal / pitch / writing / Critic, etc.)
Evaluation
Critic Agent counter-falsification + automated evaluation set + content-safety filtering
Delivery
Business plans / feasibility reports / pitch materials / Excel data packs / interactive web visualizations
Audit
Prompts / model versions / retrieved fragments / timestamps are fully logged — reproducible for investors and regulators
Multi-Agent Mesh
12 agents that specialize · collaborate · falsify each other.
A single agent easily falls into survivorship bias and collusive errors. We use task specialization + Critic counter-arguments + an evaluation-layer veto to separate 'seemingly self-consistent' conclusions from genuinely reproducible ones.
Agent 01
Industry Researcher
Breaks down track structure, boundaries, and value chain
Agent 02
Competitor Analyst
Financial proxy + strategic-intent inference
Agent 03
Regulator Watcher
Continuously tracks regulatory events and policy direction
Agent 04
Financial Modeler
Monte Carlo + sensitivity + NPV
Agent 05
Legal Architect
Capital-increase / transfer / equity-swap standard terms
Agent 06
Interview Designer
Due-diligence checklists and customer interview guides
Agent 07
Writer
Structured writing of business plans / feasibility reports
Agent 08
Designer
PDF / pitch / web interactive output
Agent 09
Critic · Adversarial
Counter-arguments; identifying survivorship bias
Agent 10
Evaluator
Hallucination detection, citation consistency
Agent 11
Orchestrator
Agent routing and arbitration
Agent 12
Auditor
Full logs and reproducibility packs
Cost Transparency · Model 03
We lay out the unit cost for you to see.
Not the market price — our true marginal cost for every 60-120 page deep report, recomputable line by line.
Single deep report · marginal cost
$35.68
- API inference (635K tokens, 70:30 in/out blended pricing)$3.05
- Embedding & vector retrieval$0.05
- Evaluation & safety filtering$0.18
- Document generation (PDF/Excel/pitch)$0.40
- Engineer internal allocation$32.00
Note: list price ranges $25K-$45K; the gross margin funds model iteration, complex due diligence, and legal-structure costs.
Single subscriber · monthly service cost
$0.97/mo
- API inference$0.85
- Storage (user workspace)$0.06
- CDN / network$0.04
- Auth / risk control$0.02
Unit cost is expected to drop ~35% at scale (Model 03 note: conservatively retained).
See our Track Analytics Engine in action.
15K Monte Carlo paths, outputting top-pick frequency, ranking robustness, and a benchmark-score heatmap.
Enter the Track Engine