Gravicode.Finance.LLM.Local 0.10.0

<p align="center"> <img src="assets/icon-512.png" width="128" alt="Finance.Net logo" /> </p>

<h1 align="center">Finance.Net</h1>

<p align="center"> <b>The quantitative finance, trading, risk, portfolio, market-intelligence and financial-AI platform for .NET 10</b><br/> Native C# with Rust accelerators and ILGPU — QuantLib, PyPortfolioOpt, pyfolio, Alphalens, Zipline/Backtrader, vectorbt, FinRL, OpenBB and FinGPT capabilities in one coherent library. </p>

<p align="center"> <i>Dibuat oleh Gravicode Studios dipimpin oleh Kang Fadhil · Made by Gravicode Studios, led by Kang Fadhil</i><br/> <a href="README.id.md">🇮🇩 Baca dalam Bahasa Indonesia</a> · <a href="docs/en/index.md">Documentation</a> · <a href="docs/id/index.md">Dokumentasi</a> </p>

<p align="center"> <img src="docs/screenshots/00-overview.png" alt="Finance.Net Gallery overview" width="900" /> </p>

Why Finance.Net

• One stack, one type system. Bars, orders, order books, curves, portfolios and backtests share the same Finance.Core types, so a market-data feed plugs straight into an indicator, a risk model, an optimiser and a broker.
• Fast and frugal. Span-based APIs, SIMD statistics, O(n) rolling windows, parallel Monte Carlo with reproducible per-thread streams, optional Rust kernels and GPU pricing — with a managed fallback for everything.
• Correct before clever. Every module is tested against textbook reference values (Hull, Haug, QuantLib examples, scipy) or independent methods, not against its own output.
• Built on the Gravicode foundation: Gravicode.Science (GraviNum linear algebra, GraviFrame data frames), HF.Net and Visual.Net, plus Gravicode.RLNet for deep reinforcement learning.

Modules

Quick start

dotnet add package Gravicode.Finance.Pricing
dotnet add package Gravicode.Finance.Risk
dotnet add package Gravicode.Finance.Portfolio
dotnet add package Gravicode.Finance.Backtesting
dotnet add package Gravicode.Finance.TimeSeries

using Finance.Core.Market;
using Finance.Pricing.Options;
using Finance.Risk;
using Finance.TimeSeries.Indicators;
using Finance.Backtesting;
using Finance.Backtesting.Strategies;

// Offline, reproducible data (swap in YahooFinanceProvider / BinanceProvider for live data).
Bar[] bars = SyntheticMarket.GeometricBrownianBars(1000, startPrice: 100, annualDrift: 0.08, annualVolatility: 0.25);
double[] close = bars.Select(b => b.Close).ToArray();

// Indicators
double[] rsi = Ta.Rsi(close, 14);
var bands = Ta.Bollinger(close, 20, 2);

// Options
var g = BlackScholes.Greeks(OptionType.Call, spot: 100, strike: 105, time: 0.5, rate: 0.04, dividendYield: 0.01, volatility: 0.25);
Console.WriteLine($"Call {g.Price:F4}  Δ {g.Delta:F4}  Γ {g.Gamma:F5}  vega/pt {g.VegaPerPoint:F4}");

// Risk
double[] returns = close.Zip(close.Skip(1), (a, b) => b / a - 1).ToArray();
var var99 = ValueAtRisk.Historical(returns, 0.99);
Console.WriteLine($"1-day VaR 99%: {var99.ValueAtRisk:P2}, ES: {var99.ExpectedShortfall:P2}");

// Backtest
var result = new BacktestEngine().Run(new SmaCrossoverStrategy(20, 50, "DEMO"), "DEMO", bars);
Console.WriteLine(result.Statistics);

More: docs/en/getting-started.md, the console tour in samples/Finance.Quickstart and 15 notebooks.

Finance.Net Gallery

An Avalonia desktop app with eleven live modules (dotnet run --project samples/Finance.Gallery).

Live Market Dashboard — Binance watchlist, 1-minute candles, EMA/RSI	Order Book Viewer — L2 ladder, depth, imbalance, sweep cost
Indicator Explorer — the whole 248-entry catalogue, live parameters	Portfolio Optimizer — 10 methods, frontier, risk contributions
VaR Explorer — five VaR methods, backtests, stress, Monte Carlo fan	Greeks Explorer — BSM vs trees vs Monte Carlo
Strategy Builder — no-code rules → backtest → generated C#	Backtesting Studio — built-in strategies, costs, walk-forward
AI Financial Assistant — tool-calling LLM over Finance.Net functions	News Intelligence — live RSS, sentiment, events
Social Sentiment — mentions, engagement-weighted sentiment, buzz

Screenshots are regenerated headlessly with dotnet run --project samples/Finance.Gallery -- --screenshot docs/screenshots.

Performance

Measured with dotnet run -c Release --project benchmarks/Finance.Benchmarks -- --quick on a 4-core 15 W laptop (Intel UHD 620 over OpenCL for the GPU column), best of 7 alternating runs:

Full table and caveats: docs/en/benchmarks.md.

Building from source

dotnet build FinanceNet.slnx                 # also runs `cargo build --release` for the Rust kernels if cargo is installed
dotnet build FinanceNet.slnx -p:SkipRustBuild=true   # managed-only build
dotnet test FinanceNet.slnx
dotnet pack -c Release                       # → artifacts/packages

Requirements: .NET 10 SDK; optional Rust 1.80+ for native kernels; optional CUDA/OpenCL driver for GPU.

Honest limitations

• The native ROCm/HIP backend has not been run on AMD hardware (none available during development); its kernel source was checked on the host. The DirectML backend runs Direct3D 12 compute shaders, not DirectML operators.
• Historical stress scenarios use stylised, rounded shocks for illustration, not regulatory use.
• Hosted FinBERT needs Hugging Face Inference credits; Gravicode.Finance.LLM.Local runs FinBERT in-process instead.
• Reddit's unauthenticated API is often rate-limited; the Gallery falls back to labelled synthetic posts.

Project files

• PLAN.md — roadmap · Progress.md — development progress
• solution-design.md — design specification (Bahasa Indonesia)

License

MIT © Gravicode Studios — Dibuat oleh Gravicode Studios dipimpin oleh Kang Fadhil.