FundFlow Sentinel — Mutual Fund Flow Forecasting & Stress Lab
Quantitative decision-support engine for Indian mutual funds with walk-forward Ridge forecasting, 3-tier redemption alerts, and stress testing
The Problem & Target Context
Indian asset management allocators and institutional analysts struggle to model macro category-level net flow shifts and redemption stress, often relying on laggy trailing AUM metrics or unverified point-in-time assumptions without temporal leakage controls.
Engineering Approach & Core Principles
Constructed an offline-first Python quantitative architecture featuring persistence baselines, walk-forward Ridge regression with expanding-window evaluation, hierarchical redemption risk thresholds, and an isolated 5-factor NAV ranking engine with zero model leakage.
What I Built & Individual Contributions
Engineered the complete quantitative pipeline: temporal feature matrix generation, Ridge walk-forward cross-validation, 3-tier redemption alerting, reference AUM stress simulation (capped at 100% outflow), and 287 automated unit/integration tests with atomic CLI state reporting.
Architecture & Data Pipeline
Raw AMFI & Market Datasets → Temporal Feature Pipeline → Ridge Walk-Forward Engine → Hierarchical Alert Scorer → Scenario Stress Lab → Streamlit Command Centre
[ AMFI Monthly Category Flows ] ──► [ Temporal Feature Matrix (Zero Leakage) ]
│
▼
┌───────────────────────────────────────────────────────────────────────────┐
│ FundFlow Sentinel Core Engine │
│ ├── Baseline Persistence Model vs Ridge Walk-Forward Regression │
│ ├── 3-Tier Alert Engine (Mild, Elevated, Severe Redemption Risk) │
│ └── Scenario Stress Lab (Historical Max, Forecast Shock, 100% Outflow Cap)│
└─────────────────────────────────────┬─────────────────────────────────────┘
│
┌───────────────────────┴───────────────────────┐
▼ ▼
[ Streamlit Command Centre & 5 Labs ] [ Independent 5-Factor NAV Ranker ]
(Local-First Offline UI) (Strictly Isolated from Macro Flows)Key Engineering Decisions & Trade-Offs
Decoupled macro category net flows from individual scheme rankings to avoid cross-contamination. Enforced temporal non-leakage by fitting scalers and Ridge estimators strictly inside expanding training folds. Capped asset-at-risk stress outflows mathematically at 100% of reference AUM.
Testing, Verification & Reproducibility
Tested on Python 3.11 with 287 automated unit and integration tests ('make test'). Includes deterministic offline CLI orchestration ('python -m fundflow.cli --offline all') with SHA-256 digest receipts in 'data/outputs/run_report.json'.
Verified Outcomes & Metrics
287 passing automated tests; sub-second offline pipeline execution; deterministic zero-leakage walk-forward validation; finance-safe outflow capping.
Verified Implementation
Verified via 287 automated tests ('make test'), atomic CLI runner with SHA-256 digest validation ('data/outputs/run_report.json'), and strict mathematical invariants enforcing temporal non-leakage and zero-MAD safeguards. Verification here represents code audit and local inspection, distinguishing functional code paths from external cloud provider accounts.
Boundaries & Known Limitations
Macro category-level flow forecasts represent econometric approximations, not guaranteed investor redemption behavior. Evaluated on monthly historical AMFI/NSE datasets; requires dated local disclosures before production institutional allocation.
Related Portfolio Projects
Explore companion systems in similar domains across AI engineering, quantitative finance, and developer infrastructure:
Collaborating on AI Engineering or Quantitative Systems?
I design and build end-to-end production software, quantitative models, and reliable AI architectures. Let’s discuss technical challenges or product collaboration.