FinceptTerminal is a modern finance application offering advanced market analytics, investment research, and economic data tools, designed for interactive exploration and data-driven decision-making in a user-friendly environment.
C++ ai-agentsalgorithmic-tradingbloomberg-terminalcppfinancefinancial-marketsfintechgood-first-issueinvestmentinvestment-researchmachine-learningopensourcepythonqtquantitative-financestock-markettrading
// src/services/ai_quant_lab/AIQuantLabService.h
#pragma once
#include "python/PythonRunner.h"
#include "services/ai_quant_lab/AIQuantLabTypes.h"
#include <QHash>
#include <QJsonArray>
#include <QObject>
#include <QPointer>
#include <QProcess>
#include <functional>
namespace fincept::services::quant {
/// Singleton service for AI Quant Lab — dispatches to 18+ Python modules.
class AIQuantLabService : public QObject {
Q_OBJECT
public:
static AIQuantLabService& instance();
// ── Generic module execution ────────────────────────────────────────────
/// Run any quant module by ID with JSON params
void run_module(const QString& module_id, const QString& command, const QJsonObject& params);
// ── Qlib core ───────────────────────────────────────────────────────────
void list_models();
void get_factor_library();
void train_model(const QJsonObject& params);
void run_backtest(const QJsonObject& params);
void optimize_portfolio(const QJsonObject& params);
// ── RL Trading ──────────────────────────────────────────────────────────
void train_rl_agent(const QJsonObject& params);
void evaluate_rl_agent(const QJsonObject& params);
// ── GS Quant ────────────────────────────────────────────────────────────
void gs_risk_metrics(const QJsonObject& params);
void gs_portfolio_analytics(const QJsonObject& params);
void gs_greeks(const QJsonObject& params);
void gs_var_analysis(const QJsonObject& params);
void gs_stress_test(const QJsonObject& params);
void gs_backtest(const QJsonObject& params);
void gs_statistics(const QJsonObject& params);
// ── Functime ────────────────────────────────────────────────────────────
void functime_check_status();
void functime_forecast(const QJsonObject& params);
void functime_anomaly_detection(const QJsonObject& params);
void functime_seasonality(const QJsonObject& params);
void functime_metrics(const QJsonObject& params);
void functime_confidence_intervals(const QJsonObject& params);
void functime_stationarity(const QJsonObject& params);
// ── Statsmodels ─────────────────────────────────────────────────────────
void sm_check_status();
void sm_ols(const QJsonObject& params);
void sm_arima(const QJsonObject& params);
void sm_stationarity_tests(const QJsonObject& params);
void sm_acf_pacf(const QJsonObject& params);
void sm_granger_causality(const QJsonObject& params);
void sm_descriptive(const QJsonObject& params);
// ── Fortitudo ───────────────────────────────────────────────────────────
void fort_check_status();
void fort_portfolio_metrics(const QJsonObject& params);
void fort_covariance_matrix(const QJsonObject& params);
void fort_mean_variance_optimize(const QJsonObject& params);
void fort_mean_cvar_optimize(const QJsonObject& params);
void fort_efficient_frontier(const QJsonObject& params);
void fort_exp_decay_probabilities(const QJsonObject& params);
// ── GluonTS ─────────────────────────────────────────────────────────────
void gluon_check_status();
void gluon_probabilistic_forecast(const QJsonObject& params);
void gluon_quantile_forecast(const QJsonObject& params);
void gluon_distribution_fit(const QJsonObject& params);
void gluon_evaluate_forecast(const QJsonObject& params);
void gluon_seasonal_naive(const QJsonObject& params);
// ── CFA Quant ───────────────────────────────────────────────────────────
void cfa_trend_analysis(const QJsonObject& params);
void cfa_stationarity_test(const QJsonObject& params);
void cfa_arima(const QJsonObject& params);
void cfa_ml_prediction(const QJsonObject& params);
void cfa_pattern_discovery(const QJsonObject& params);
void cfa_bootstrap(const QJsonObject& params);
void cfa_data_quality(const QJsonObject& params);
// ── Deep Agent (LangGraph multi-agent) ──────────────────────────────────
/// `params`: task, agent_type, thread_id, config (LLM block). The payload —
/// LLM key included — goes to Python on stdin, never argv. Without a
/// `config` (MCP / LLM-originated calls) the ai_quant_lab LLM profile is used.
void run_deep_agent(const QJsonObject& params);
// ── RD-Agent (autonomous factor/model research) ──────────────────────────
// Long tasks (start_* / resume) run as one child process per task, owned by
// this service: progress streams back as JSON lines, status/results are
// served from memory, stop = kill. They live for this app session only.
void rd_agent_check_status();
void rd_agent_start_factor_mining(const QJsonObject& params);
void rd_agent_start_model_optimization(const QJsonObject& params);
void rd_agent_start_quant_research(const QJsonObject& params);
void rd_agent_get_task_status(const QString& task_id);
void rd_agent_get_discovered_factors(const QString& task_id);
void rd_agent_get_optimized_model(const QString& task_id);
void rd_agent_list_tasks(const QString& status_filter = {});
void rd_agent_stop_task(const QString& task_id);
void rd_agent_resume_task(const QString& task_id, const QJsonObject& config = {});
void rd_agent_start_ui();
void rd_agent_start_mcp_server(int port = 18765);
void rd_agent_mcp_status();
// ── Advanced ────────────────────────────────────────────────────────────
void list_advanced_models();
void create_advanced_model(const QJsonObject& params);
void train_advanced_model(const QJsonObject& params);
// ── Factor Discovery ────────────────────────────────────────────────────
void factor_get_library();
void factor_get_data(const QJsonObject& params);
void factor_get_calendar(const QJsonObject& params);
void factor_get_instruments();
// ── Model Library ───────────────────────────────────────────────────────
void model_list();
void model_train(const QJsonObject& params);
void model_backtest(const QJsonObject& params);
void model_check_status();
// ── Live Signals ────────────────────────────────────────────────────────
void signals_get_data(const QJsonObject& params);
void signals_get_factor_analysis(const QJsonObject& params);
void signals_get_feature_importance(const QJsonObject& params);
// ── HFT ─────────────────────────────────────────────────────────────────
void hft_create_orderbook(const QJsonObject& params);
void hft_snapshot(const QJsonObject& params);
void hft_market_making_quotes(const QJsonObject& params);
void hft_detect_toxic(const QJsonObject& params);
void hft_execute_order(const QJsonObject& params);
// ── Meta Learning ───────────────────────────────────────────────────────
void meta_list_models();
void meta_run_selection(const QJsonObject& params);
void meta_create_ensemble(const QJsonObject& params);
void meta_tune_hyperparameters(const QJsonObject& params);
void meta_get_results();
// ── Online Learning ─────────────────────────────────────────────────────
void online_list_models();
void online_create_model(const QJsonObject& params);
void online_train(const QJsonObject& params);
void online_predict(const QJsonObject& params);
void online_performance(const QJsonObject& params);
// ── Rolling Retraining ──────────────────────────────────────────────────
void rolling_create_schedule(const QJsonObject& params);
void rolling_execute_retrain(const QJsonObject& params);
void rolling_list_schedules();
void rolling_preview_tasks(const QJsonObject& params);
void rolling_delete_schedule(const QJsonObject& params);
// ── Feature Engineering ─────────────────────────────────────────────────
void feature_compute(const QJsonObject& params);
void feature_select_by_ic(const QJsonObject& params);
void feature_evaluate_expression(const QJsonObject& params);
// ── Portfolio Optimization ──────────────────────────────────────────────
void portopt_run(const QString& method, const QJsonObject& params);
// ── Factor Evaluation ───────────────────────────────────────────────────
void evaluation_ic(const QJsonObject& params);
void evaluation_report(const QJsonObject& params);
void evaluation_risk_metrics(const QJsonObject& params);
// ── Strategy Builder ────────────────────────────────────────────────────
void strategy_create(const QString& type, const QJsonObject& params);
void strategy_portfolio_metrics(const QJsonObject& params);
void strategy_list();
// ── Data Processors ─────────────────────────────────────────────────────
void dataproc_list_processors();
void dataproc_create_pipeline(const QJsonObject& params);
void dataproc_process_data(const QJsonObject& params);
// ── Quant Reporting ─────────────────────────────────────────────────────
void reporting_check_status();
void reporting_ic_analysis(const QJsonObject& params);
void reporting_model_performance(const QJsonObject& params);
void reporting_cumulative_returns(const QJsonObject& params);
void reporting_risk_report(const QJsonObject& params);
void reporting_factor_quantiles(const QJsonObject& params);
signals:
void result_ready(QString module_id, QString command, QJsonObject data);
void error_occurred(QString module_id, QString message);
/// Emitted during a streaming training run (currently: rl_trading/train).
void training_progress(QString module_id, int step, int total, double reward_mean, double loss);
void training_log(QString module_id, QString line, bool is_stderr);
private:
explicit AIQuantLabService(QObject* parent = nullptr);
Q_DISABLE_COPY(AIQuantLabService)
void run_python(const QString& script, const QStringList& args, const QString& module_id, const QString& command,
const python::PythonRunner::RunOptions& opts = {});
void run_python_cached(const QString& script, const QStringList& args, const QString& module_id,
const QString& command, int ttl_sec);
/// run_module() for module "deep_agent": the LangGraph deep agent
/// (execute_task) and every RD-Agent command, each to its implementation.
void run_deep_agent_command(const QString& command, const QJsonObject& params);
/// The LLM block for an agent run: the caller's `config` when it carries a
/// provider, else the ai_quant_lab profile resolved from LlmProfileRepository.
QJsonObject agent_llm_config(const QJsonObject& params) const;
// ── RD-Agent task registry ──────────────────────────────────────────────
struct RdTask {
QString id;
QString kind; // factor_mining | model_optimization | quant_research
QString status; // running | completed | failed | stopped
QJsonObject params; // task parameters WITHOUT the LLM credentials
QString workspace;
double progress = 0.0; // 0..1
int iterations_completed = 0;
int max_iterations = 0;
QString current_step;
QJsonValue best_ic;
QJsonValue best_sharpe;
QJsonArray factors;
QJsonArray models;
QString error;
qint64 started_ms = 0;
qint64 finished_ms = 0;
bool stop_requested = false;
QPointer<QProcess> proc;
};
void rd_start(const QString& command, const QString& kind, const QJsonObject& params);
/// Spawn `cli.py run_task` for `task_id`; `resume_from` = checkpoint path or empty.
bool rd_spawn(const QString& task_id, const QJsonObject& llm_config, const QString& resume_from,
QString* error_out);
void rd_handle_line(const QString& task_id, const QByteArray& line);
QJsonObject rd_status_json(const RdTask& t) const;
int rd_running_count() const;
/// result_ready("deep_agent", command, data), queued so callers that connect
/// after calling (and screens that update their status text after calling)
/// see it.
void rd_emit(const QString& command, const QJsonObject& data);
static constexpr int kListTtlSec = 5 * 60;
static constexpr int kMaxConcurrentRdTasks = 2;
// Module ID → script mapping
QHash<QString, QString> script_map_;
QHash<QString, RdTask> rd_tasks_;
};
} // namespace fincept::services::quant