Production-style, enterprise-grade Mutual Fund Analytics Platform demonstrating end-to-end Data Engineering, Advanced Analytics, Cloud Architecture, and Executive BI Reporting.
Built by: Sumit Kumar Prajapat | GitHub: Skpkush
Status: β Complete. The Azure pipeline ran end to end during the 15-day Azure trial (see screenshot). The app now runs from Supabase or, when that is offline, a data snapshot in this repo.
π΄ Live Demo: mf-analytics-platform.streamlit.app
| Metric | Value | Basis |
|---|---|---|
| Live schemes | 8,620 across 50 AMCs | NAV dated in the snapshot's final month (2026-05) |
| Schemes listed by AMFI | 14,368 across 51 AMCs | Includes closed and matured schemes |
| Daily NAV rows | 9,706,307 | 12,810 schemes, 2021-06-02 to 2026-05-31 |
| ETFs + benchmark indices | 11 + 5 | The series behind the app's risk and trend pages |
| Star schema | 5 dimension + 4 fact tables | Dim_Fund holds 14,384 rows: listed schemes + ETFs + indices |
| Power BI | 4 pages, 37 DAX measures | Screenshots show 4,961 funds / 52 AMCs, the set with metrics when it was built |
| Streamlit app | 7 pages | Runs from a repo snapshot if the database is down |
| Moderate-risk shortlist | 179 β 49 β 5 funds | Average drawdown -3.62% vs -4.83% for peers, 11 test windows |
| Trend model check | 7.77% vs 5.8% error at 90 days | Prophet vs a no-change baseline; 32.7% coverage from its 80% band |
| Investor data | 500 investors, 35,280 transactions | Synthetic, used for the Investor Analytics demo page only |
Every number above is generated from the data into docs/metrics.json, and a test fails if this table drifts from it. Quote project numbers from that file only.
Asset Management Companies, Wealth Advisors, and FinTech firms need a unified analytics platform that ingests multi-source mutual fund data, computes risk-adjusted performance metrics, segments investors, explores NAV trends, and surfaces executive-level insights β all in a scalable cloud architecture.
This platform delivers exactly that.
Which 5 funds should an advisor suggest to a moderate-risk client?
Starting from 179 Direct-Growth schemes in seven suitable categories, the screen applies a risk gate relative to the Nifty 50, then picks the best-scoring basket under diversification limits. A walk-forward test shows what the method does and does not deliver:
| Approach | Avg 1Y return | Avg worst fall |
|---|---|---|
| This shortlist | 9.63% | β3.62% |
| Same-category peers | 9.52% | β4.83% |
| Top 5 by last year's return | 7.96% | β6.16% |
The finding: the edge is smaller falls, not higher returns. Return matched peers, while the worst fall was shallower in all 11 overlapping test windows. The full answer, method and limits are in docs/analysis/moderate_risk/, and in the app's Advisor Shortlist page. This is a historical analysis, not investment advice.
βββββββββββββββββββ βββββββββββββββββββ βββββββββββββββββββ
β Data Sources βββββΆβ Azure Blob βββββΆβ Azure Data β
β Yahoo Finance β β (Raw Layer) β β Factory β
β AMFI India β βββββββββββββββββββ β (β
Succeeded) β
β Kaggle β βββββββββββββββββββ
βββββββββββββββββββ β
βΌ
βββββββββββββββββββ βββββββββββββββββββ βββββββββββββββββββ
β Streamlit App ββββββ Power BI ββββββ Azure SQL DB β
β(Streamlit Cloud)β β 4-Page β β Star Schema β
βββββββββββββββββββ β Dashboard β β 5 Dim + 4 Fact β
β² βββββββββββββββββββ βββββββββββββββββββ
β β
β βΌ
β βββββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ Python ML β
β Prophet NAV β
β Trend β
βββββββββββββββββββ
Data flow: Raw multi-source data lands in Azure Blob β Azure Data Factory orchestrates ingestion β transformation/cleaning in Python β load into a star schema (5 dimension + 4 fact tables) on Azure SQL Database (PostgreSQL locally) β 37 DAX measures power a 4-page Power BI dashboard β Prophet separates NAV trend from noise β Streamlit surfaces client-facing analytics.
Full diagram: docs/architecture/architecture.md
- Dimensions:
Dim_Date,Dim_AMC,Dim_Category,Dim_Fund,Dim_Investor - Facts:
Fact_NAV,Fact_Transactions,Fact_SIP,Fact_Returns
DDL: scripts/sql/ddl/ Β· Views: scripts/sql/views/ Β· Procs: scripts/sql/procs/
| Layer | Technology |
|---|---|
| Cloud | Azure (Blob Storage, SQL Database, Data Factory, Functions, Key Vault) |
| Database | Azure SQL Database (cloud) + PostgreSQL 15 (local/fallback) |
| ETL | Azure Data Factory + Python (pandas 2.x, sqlalchemy) |
| Analytics | Python (numpy, scipy, statsmodels) |
| ML | Prophet (NAV trend exploration) |
| BI | Power BI Desktop β 4 pages, 37 DAX measures |
| App | Streamlit (deployed on Streamlit Community Cloud) |
| Orchestration | ADF pipelines + Azure Functions |
| Version Control | Git + GitHub |
π₯ Download: Power BI Dashboard (.pbix) (Git LFS) Β· πΌοΈ Screenshots: docs/screenshots/
- Executive Overview β AUM, top funds, market summary, KPIs
- Fund Performance Analytics β CAGR, rolling returns, benchmark comparison
- Investor Analytics β SIP trends, segmentation, retention (synthetic demo data: 500 investors)
- Risk & Volatility β Sharpe heatmaps, drawdown, risk-return scatter
Page 2 β Fund Performance Analytics

Azure Data Factory Pipeline β Succeeded

- CAGR (Compound Annual Growth Rate)
- Rolling Returns (1Y, 3Y, 5Y)
- Alpha, Beta (vs Nifty 50 benchmark)
- Sharpe Ratio, Sortino Ratio, Treynor Ratio
- Standard Deviation, Volatility
- Maximum Drawdown
- Information Ratio
NAV Trend Exploration (Prophet) separates a fund's trend from noise and shows its direction and how far the NAV sits from it. It is not a forecast: in a walk-forward backtest across 11 ETFs, Prophet lost to a no-change baseline at 30, 60 and 90 days, and its 80% band held only about a third of the time.
python scripts/ml/nav_trend.py --fund-code NIFTYBEES.NS # trend scenario for one fund
python scripts/ml/backtest_nav_trend.py # walk-forward backtest
streamlit run streamlit/app.py # interactive app# Clone repo
git clone https://github.com/Skpkush/mf-analytics-platform.git
cd mf-analytics-platform
# Create virtual environment
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Configure environment variables
cp .env.example .env # Windows: copy .env.example .env
# Edit .env with your Azure / PostgreSQL credentialspython scripts/ingestion/fetch_yahoo_finance.py
python scripts/ingestion/fetch_amfi_nav.py# Clean and validate NAV / transaction history
python scripts/transformation/clean_nav.py
python scripts/transformation/clean_transactions.py
# Create the schema (DDL), then load dimensions and facts
python scripts/etl/run_ddl.py
python scripts/etl/load_dimensions.py
python scripts/etl/load_facts.pypython scripts/analytics/metrics_returns.py # CAGR, rolling returns
python scripts/analytics/metrics_risk.py # volatility, drawdown
python scripts/analytics/metrics_risk_adjusted.py # Sharpe, Sortino, Treynor, alpha/beta
python scripts/analytics/metrics_market.py # market/AUM aggregatespython scripts/ml/nav_trend.py --list # funds with enough history
python scripts/ml/nav_trend.py --fund-code NIFTYBEES.NS # trend plus 30/60/90-day scenario
python scripts/etl/build_app_snapshot.py # offline data for the app
python scripts/ml/backtest_nav_trend.py # is the trend a usable prediction?
python scripts/analytics/moderate_risk_shortlist.py # the 5-fund business question
python scripts/analytics/build_metrics_manifest.py # refresh docs/metrics.json + README table- Power BI: open
powerbi/mf_analytics_dashboard_p4.pbix(apply themepowerbi/theme_mf_analytics.json). - Streamlit:
streamlit run streamlit/app.pyβ openhttp://localhost:8501(no database needed, it falls back to the repo snapshot; deploy notes instreamlit/README.md).
Cloud path: the Azure Data Factory pipeline orchestrates the ingestion β blob β SQL load in the cloud (pipeline run Succeeded β see screenshot above). Run the cloud ETL via
scripts/etl/run_azure_etl.py/scripts/etl/trigger_adf_pipeline.py.
mf-analytics-platform/
βββ azure/
β βββ adf_pipelines/ # ADF JSON pipeline definitions
β βββ alerts/ # ADF failure alert configurations
β βββ arm_templates/ # Infrastructure as code
β βββ datasets/ # ADF dataset definitions
β βββ functions/
β β βββ fn_compute_daily_metrics/ # Azure Function daily refresh
β βββ pipelines/ # ADF pipeline JSON definitions
β βββ triggers/ # ADF schedule triggers
βββ data/
β βββ raw/ # AMFI + Yahoo source parquet
β βββ processed/ # Cleaned, transformed data
β βββ app/ # Offline snapshot the Streamlit app can run from
βββ docs/
β βββ analysis/ # Moderate-risk shortlist + trend backtest
β βββ architecture/ # Architecture diagrams + Mermaid
β βββ case_study/ # Case study PDF (planned)
β βββ screenshots/ # Dashboard + pipeline screenshots
β βββ daily_log.md # Day-by-day build log
β βββ data_dictionary.md # Schema + field definitions
β βββ metrics.json # Numbers of record
βββ notebooks/
β βββ 01_data_exploration.ipynb # EDA notebook
βββ powerbi/
β βββ mf_analytics_dashboard_p4.pbix # 4-page dashboard (Git LFS)
β βββ theme_mf_analytics.json # Navy theme file
βββ scripts/
β βββ ingestion/ # AMFI + Yahoo Finance data acquisition
β βββ transformation/ # Cleaning, validation, feature engineering
β βββ etl/ # DDL run, dimension/fact load, Azure ETL
β βββ analytics/ # Metrics, advisor shortlist, numbers of record
β βββ ml/ # Prophet NAV trend + walk-forward backtest
β βββ sql/
β βββ ddl/ # CREATE TABLE scripts (star schema)
β βββ views/ # Analytical views (vw_fund_performance)
β βββ procs/ # Stored procedures
βββ streamlit/
β βββ .streamlit/ # Theme + secrets template
β βββ pages/ # 6 app pages (Explorer, Trend, Risk, SIP, Portfolio, Advisor)
β βββ utils/ # DB, metrics, charts utilities
β βββ app.py # Home page + KPI cards
β βββ Dockerfile # Container definition
β βββ docker-compose.yml # VPS deployment config (alternative)
β βββ requirements.txt # App-only dependencies
β βββ README.md # Deploy instructions
βββ tests/
β βββ conftest.py # Pytest fixtures
β βββ test_data_quality.py # Data quality unit tests
β βββ test_analysis_and_numbers.py # Shortlist logic + number checks
βββ .env.example # Environment variable template
βββ .gitignore # Secrets + venv excluded
βββ CLAUDE.md # Claude Code project context
βββ PROJECT_PLAN.md # Week-by-week execution plan
βββ requirements.txt # Full Python dependencies
βββ README.md
- Architecture
- Project Plan
- Data Dictionary
- KPI Definitions
- Numbers of record
- Moderate-risk shortlist
- Trend backtest
- Case Study PDF
MIT License β see LICENSE file.
Sumit Kumar Prajapat Data Analyst | Analytics Engineer | π§ sumitkprajapat29@gmail.com π GitHub
Last updated: November 2026


