QuantPath

MFE admission prediction engine — GPBoost mixed-effects model on 13,100+ records across 31 programs. Profile scoring, prerequisite matching, and data-driven school ranking.

-orange.svg)

Why QuantPath

Consulting services charge $4,500--10,000 per school. QuantPath provides the same analysis for free -- multi-dimensional profile scoring, admission probability with confidence intervals, and personalized gap analysis across 15 focused MFE programs (31 total in database).

Community calibration: Applicants and maintainers also post anonymized profiles, quantpath predict output, and actual results to GitHub issues labeled data-contribution. That stream is used to spot systematic gaps (e.g., same-university different-degree programs, waitlists) that offline metrics alone do not show.

$ quantpath predict --profile my_profile.yaml
QuantPath Admission Prediction (v2 Model)   Sample Applicant | Top 30 University | GPA 4.0 | International
Program                P(admit)   Category   ─────────────────────────────────────────────   Princeton MFin           18%      reach   Baruch MFE               22%      reach   Berkeley MFE             35%      reach   CMU MSCF                 38%      reach   MIT MFin                 25%      reach   Columbia FE              42%      target   Yale AM                  15%      reach   Stanford MCF             20%      reach   UChicago MSFM            55%      target   NYU Courant              52%      target   Columbia MSFE            48%      target   Cornell MFE              58%      target   Columbia MAFN            62%      target   NYU Tandon MFE           72%      safety   GaTech QCF               75%      safety
15 programs evaluated (Tier 0 + Tier 1)

Data

| Dataset | Records | Source | |---------|---------|--------| | Admission records | 13,100+ | GradCafe, QuantNet, Reddit, 1point3acres, ChaseDream (accepted/rejected/waitlisted) | | LinkedIn alumni | 930 | 20 MFE programs (employer, undergrad school, graduation year) | | Program database | 31 | QuantNet 2026 Rankings + official sites (prerequisites, deadlines, salaries) |

Model

v1 (primary): Per-program logistic regression on GPA + GRE Quant with bias correction and profile adjustments for undergrad tier, internship quality, research, and major relevance. 27 trained models covering all 15 focused programs.

v2 (fallback): GPBoost -- LightGBM gradient boosting with per-program random intercepts. Trained on 11,100+ labeled records, 13 features, 31 programs. AUC 0.723, Brier 0.206 (5-fold CV). Used for programs without a v1 model.

Bias correction: Self-reported data has survivor bias (65% accept rate in data vs 4-30% real). The model replaces the biased intercept with logit(r) where r is the official acceptance rate, preserving learned feature slopes (King & Zeng 2001).

Beyond CV metrics: Labeled data-contribution issues add case-level checks against real outcomes. On strictly comparable rows (same focused program id as in quantpath predict, clear admit vs reject, binary threshold at P=0.5), historically reviewed batches align with realized decisions ~75–76% of the time—illustrative, not a guarantee: many posts mix non-focused programs, waitlists, or degree names that do not map 1:1 to a single YAML id.

Features

| Command | Description | |---------|-------------| | quantpath predict | Primary entry point -- P(admit) for 15 focused programs, reach/target/safety (no transcript needed) | | quantpath evaluate | Profile assessment -- 5-dimension score (37 sub-factors) with gaps and strengths | | quantpath list | Personalized reach/target/safety school list with P(admit) + CI | | quantpath match --program cmu-mscf | Prerequisite match for a specific program | | quantpath gaps | Priority-ranked gaps with recommended actions | | quantpath optimize | Top courses to take for maximum profile improvement | | quantpath compare --programs cmu-mscf,baruch-mfe | Side-by-side program comparison | | quantpath roi | Tuition, salary, NPV, payback period per program | | quantpath timeline | Month-by-month application action plan | | quantpath portfolio --n-schools 8 --budget 1500 | Optimize school list under budget constraints | | quantpath whatif --gpa 3.95 --gre 170 | See how improvements change P(admit) | | quantpath tests | GRE/TOEFL requirements across all programs | | quantpath programs | Full program database with rankings and stats | | quantpath interview | Practice questions by category and difficulty | | quantpath contribute-upload | Upload locally saved contribution data to GitHub |

AI Tools

| Tool | Description | |------|-------------| | python tools/advisor.py --profile X.yaml | Full AI advisory report (Claude-powered) | | python tools/parse_profile.py --input resume.txt | Resume/transcript → profile YAML | | python tools/trainmodelv2.py | Train GPBoost model from admission data | | python tools/scrape_1p3a.py manual | Parse Chinese forum posts into structured data | | CLAUDE.md | Open in Claude Code for interactive MFE advising |

Quick Start

Step 1 — Install and predict (< 2 minutes)

git clone https://github.com/MasterAgentAI/QuantPath.git
cd QuantPath && pip3 install -e .
Interactive — answers a few questions, no file needed
quantpath predict

This gives you P(admit) for 15 programs, classified as reach/target/safety. Your profile is automatically saved to profiles/ for reuse.

Step 2 — Deeper analysis (optional, add courses)

The saved profile works for predict, but for course-level evaluation (5-dimension score, gap analysis, course recommendations), add your transcript:

# Edit the profile saved in Step 1 and add your courses
(see examples/sample_profile.yaml for the full format)
vim profiles/your_name.yaml
Then run detailed analysis
quantpath evaluate --profile profiles/your_name.yaml
quantpath gaps     --profile profiles/your_name.yaml
quantpath optimize --profile profiles/your_name.yaml

AI-powered advisory report (optional)

pip3 install anthropic
export ANTHROPICAPIKEY=your_key
python tools/advisor.py --profile profiles/your_name.yaml --save report.md

Programs (15 Focused — Tier 0 + Tier 1)

Tier 0 (elite, <10% acceptance or unique positioning):

| # | Program | University | Class | Rate | Avg GPA | Salary | |---|---------|-----------|-------|------|---------|--------| | 1 | MFin | Princeton | 44 | 5% | 3.90 | $160K | | 2 | MFE | Baruch College | 20 | 4% | 3.84 | $179K | | 3 | MFE | UC Berkeley | 76 | 17% | 3.80 | $154K | | 4 | MSCF | Carnegie Mellon | 108 | 17% | 3.86 | $134K | | 5 | MFin | MIT Sloan | 126 | 8% | 3.80 | $140K | | 6 | MSFE | Columbia FE (Econ) | 25 | 5% | 3.90 | $150K | | 7 | AM | Yale | 3 | 5% | 3.90 | $145K | | 8 | MCF | Stanford | 10 | 5% | 3.90 | -- |

Tier 1 (highly competitive, strong placement):

| # | Program | University | Class | Rate | Avg GPA | Salary | |---|---------|-----------|-------|------|---------|--------| | 9 | MSFM | UChicago | 118 | 22% | 3.80 | $124K | | 10 | MathFin | NYU Courant | 37 | 23% | 3.85 | $126K | | 11 | MSFE | Columbia (IEOR) | 136 | 13% | 3.90 | $138K | | 12 | MFE | Cornell | 53 | 21% | 3.80 | $115K | | 13 | MAFN | Columbia (Math) | 101 | 22% | 3.80 | $123K | | 14 | MFE | NYU Tandon | 154 | 38% | 3.83 | $107K | | 15 | QCF | Georgia Tech | 99 | 30% | 3.75 | $115K |

Full 31-program database with deadlines, prerequisites, essay requirements, and interview formats in data/programs/.

Course Categories

37 categories across 4 academic dimensions, aligned with MFE program prerequisites.

Architecture

QuantPath/
├── core/                    # Evaluation engine (pure Python + PyYAML)
│   ├── profile_evaluator    # 5-dimension scoring (37 sub-factors)
│   ├── school_ranker        # Reach/target/safety + GPBoost v2 integration
│   ├── lr_predictor         # v1 LR + v2 GPBoost inference (graceful fallback)
│   ├── list_builder         # Portfolio optimization with geographic diversity
│   ├── gap_advisor          # Gap analysis with action recommendations
│   ├── course_optimizer     # Course impact optimization
│   ├── roi_calculator       # Financial ROI analysis
│   └── calibrator           # Model calibration from real admission data
├── cli/main.py              # 15+ CLI commands
├── web/app.py               # Streamlit dashboard (6 pages)
├── tools/
│   ├── trainmodelv2.py    # GPBoost training pipeline
│   ├── train_model.py       # v1 sklearn LR training
│   ├── advisor.py           # Claude-powered advisory report
│   ├── parse_profile.py     # Resume/transcript → YAML parser
│   ├── scrape_1p3a.py       # 1point3acres scraper + regex parser
│   └── collect_data.py      # GradCafe + QuantNet data collection
├── scripts/
│   ├── preparetrainingdata.py  # Data cleaning + feature matrix generation
│   └── collect_multidim.py       # Multi-source data collection pipeline
├── data/
│   ├── programs/            # 31 program YAML files
│   ├── admissions/          # 13,100+ records (CSV + JSON)
│   └── models/              # 27 LR models (primary) + GPBoost v2 (.bin + .json)
└── tests/                   # 465 tests, <1s runtime

Contributing

• Add/update program data — PRs to data/programs/ with updated deadlines or new programs
• Submit admission results — see Contributing Your Data below
• Report data errors — Open issues for outdated program information
• Feature requests — Tell us what would help your application process

Contributing Your Data

After running quantpath predict, you'll be asked whether to contribute your anonymized data. This is the #1 way to improve prediction accuracy for everyone.

Option A — Automatic (recommended):

quantpath predict --profile my_profile.yaml 
Answer "y" when prompted → auto-submits to GitHub

Option B — Browser (no GitHub CLI needed):

If you don't have gh installed, the CLI will save your data locally and offer to open a pre-filled GitHub issue page in your browser. Just click "Submit new issue".

Option C — Upload later:

# If data was saved locally (e.g. gh was not available), submit it later: quantpath contribute-upload

Privacy: All data is anonymized before submission — university names are replaced with tiers (e.g. "US T30"), company names with categories (e.g. "top quant"). You choose per-field whether to share the original value or the anonymized version.

After you submit: Contributions land as GitHub issues (same data-contribution label). They are used for calibration discussions, future feature and tier tweaks, and documentation—not for training the public model until explicitly merged into the licensed dataset pipeline.

License

This project uses a multi-license structure to balance openness with protection:

| Component | License | Commercial Use | |-----------|---------|----------------| | Source code | AGPL-3.0 | Allowed if you release your source code (including SaaS) | | Admission data & models | CC BY-NC-SA 4.0 | Not permitted without written agreement | | Program database | CC BY-SA 4.0 | Allowed with attribution |

In plain language:

• Personal use, academic research, and learning — fully permitted, no restrictions.
• Contributing and forking — welcome under the same license terms.
• Building a commercial product or SaaS — the AGPL requires you to open-source your entire codebase. The data and models cannot be used commercially without a separate license.
• Redistribution of data — allowed for non-commercial purposes with attribution and ShareAlike.

See LICENSE for the complete terms and NOTICE for attribution details.

QuantPath is a tool to assist with MFE application planning. All program data is sourced from official websites and QuantNet rankings. Always verify on official program websites before applying.