Kaiso
PeiPaoLab's in-house admissions matching engine · anchored in real admissions outcomes · upgraded each cycle
About Kaiso
Kaiso is PeiPaoLab's fully in-house admissions matching engine. Anchored in real admissions cases and re-calibrated each application cycle, it gives every applicant an auditable Reach / Match / Safety breakdown.
Fully in-house
Profile pipeline, matching logic, policy calibration, and explanation layer all built end-to-end.
Anchored in admissions
We rely on real historical offer distributions, not weighted averages of public rankings.
Cycle-aware tuning
Hard policy constraints for the current cycle enter inference directly, recalibrated every quarter.
Four-stage pipeline
From applicant inputs to Reach / Match / Safety output, every Kaiso step is broken into auditable nodes.
Profile decomposition
Break the applicant's academics, activities, recommendations, and intent into fine-grained items, accounting for school- and identity-specific context.
Peer sample matching
Find peer samples with similar conditions in the 8-cycle real admits pool, bucketed by identity, school type, region, and intended major.
Policy calibration
Hard policy constraints for the current cycle feed directly into tier adjustments, re-calibrated quarterly.
P-rank scoring
13 scoring components are parsed independently — every tier landing can be traced back to which item pushed it up or pulled it down.
Specific weights, thresholds, and bucketing strategies are PeiPaoLab core IP and are not disclosed.
Data sources and cadence
Data pool
- 44 core US schools × 8 cycles of real admissions samples
- High-density Chinese-American high-school baseline pool, across regions and curricula
- Peer samples stratified by application intent — avoids global-average distortion
Update cadence
- Fully retrained each year as a new admissions cycle closes
- Policy variable layer recalibrated quarterly
- Each report is version-stamped — conclusions are traceable
Iteration timeline
Kaiso is not a one-shot product — admissions policies shift every quarter, data grows every year, and inference logic is recalibrated alongside. Each upgrade is version-stamped, so any conclusion in a report is traceable to a specific version.
- v1.92026-05-04Institution matching layer v2 + Pathway decision engine v2
- Institution matching layer v2 (top-tier multi-major adaptation dim introduced)
- Pathway decision engine v2 (trigger precision refinement)
- Cross-product alignment (Sprint Pack v13 / Positioning v11 / Semester v8)
- v1.82026-05-02Region cohort expansion · CN school districts + EC type breakdown
- Regional recommendation baseline v1.1 → v1.2 (broader coverage)
- P-rank region cohort dim v1 → v1.1 (finer granularity)
- EC type component v1 → v1.2 (deeper category breakdown)
- v1.72026-05-02P-rank trifecta precision upgrade · course rigor + standardized trend + EC breadth
- P-rank scoring dim component v2.1 → v2.2
- Course rigor signal channel v1 launched (AP/IB/A-Level score breakdown)
- Standardized trend signal channel v1 → v2 (bidirectional rise/decline weighting)
- EC breadth signal channel v1 → v2 (precision across 8 activity categories)
- v1.62026-05-02Data sources expanded · diversity + trend + auto-detection
- Admissions context channel v2 → v3
- EC breadth dim v1 launched
- Standardized trend signal channel v1 launched
- High school type auto-detection v1 launched
- Multi-scenario prompt co-evolution rule established
- v1.52026-05-01Reasoning stability · trigger surface expanded
- Judgment provenance layer v3 → v3.1
- Multi-track intent modeling v1 → v2
- Regional recommendation baseline v1 → v1.1
- Output layer v1 launched
- v1.42026-05-01Visualization layer + multi-role aggregation launched
- P-rank visualization card component v1 launched
- Multi-role aggregation layer v1 launched
- Risk-signal sub-module v1 launched
- Overall aggregation card v1 launched
- v1.32026-05-01Scoring core upgrade · broader judgment basis
- P-rank scoring dim component v2 → v2.1
- Holistic judgment sub-module v1 launched
- Admissions context channel v1 → v2
- Judgment provenance layer v2 → v3
- Diverse-background modeling v1 launched
- v1.22026-04-25Scoring core major upgrade · multi-track intent modeling launched
- P-rank scoring dim component v1.1 → v2
- Judgment provenance layer v1 → v2
- Multi-track intent modeling v1 launched
- Identity & institutional advantage engine v1 launched
- Regional recommendation baseline v1 launched
- Standardized-policy hard routing v1 launched
- v1.1.12026-04-23Scoring core minor iteration · multi-curriculum generalization
- P-rank scoring dim component v1 → v1.1
- Tier consistency engine v1 launched
- Curriculum equivalence mapping v1 launched
- Rec-letter signal channel v1 launched
- v1.12026-04-22Judgment provenance layer launched · first-order variables expanded
- Holistic signal channel v1 launched
- Judgment provenance layer v1 launched
- Admissions context channel v1 launched
- Data-timing anchoring model v1 launched
- Extracurricular sub-module v1 launched
- v1.02026-04-15Kaiso engine initial release
- Kaiso engine v1.0 initial release
- P-rank scoring dim component v1 launched
- 4-stage pipeline baseline established
- Sprint Pack judgment chain v1 launched
- Positioning Diagnosis judgment chain v1 launched
- Growth Profile judgment chain v1 launched
Specific feature lists, inference weights, and data pool composition are PeiPaoLab core IP; release notes only disclose capability dimensions.
Run Kaiso for free
Spend two minutes entering the applicant's basics, and Kaiso returns a Reach / Match / Safety tier breakdown with policy assumption notes. No email required.
Start the quick assessment→