book

Financial AI in Practice

Name: Financial AI in Practice
Author: Taehun Kim
ISBN: 9781633435391

by Taehun Kim

October 2026

464 pages

13h 48m

English

Manning Publications

Read now

Unlock full access

Financial AI in Practice
copyright
contents
dedication
preface
acknowledgments
about this book
about the author
about the cover illustration
Part 1 Foundations of financial AI

1 Introducing AI in finance
1.1 AI in the context of finance1.2 The new imperative: Why AI is no longer optional1.3 How AI is transforming finance1.3.1 Managing risk and ensuring compliance1.3.2 Extracting market intelligence for better investments1.3.3 Enhancing customer experiences1.3.4 Improving operational efficiency1.4 Core building blocks of financial AI applications1.4.1 Data asset layer1.4.2 Modeling layer1.4.3 Strategy and monitoring layer1.4.4 Application layer1.5 How the layers work together: A credit scoring example1.5.1 The data asset layer in credit scoring1.5.2 The modeling layer in credit scoring1.5.3 The strategy and monitoring layer in credit scoring1.5.4 The application layer in credit scoring1.6 Tools, data, and infrastructure you’ll need1.6.1 Programming language and frameworks1.6.2 Data and computing environment1.6.3 Orchestration and monitoring (optional)1.6.4 IDE and code access1.6.5 Costs, licensing, and flexibility
2 From model to money: A strategist’s guide to real-world financial AI
2.1 The survival guide for financial AI projects2.1.1 The hard truth: Why good models die2.1.2 Who this chapter is for2.2 The project spark: Launching Alpha Digest on a trading platform2.2.1 The business problem: Information overload at AlphaStream2.2.2 The AI solution: A personalized Alpha Digest2.3 The strategic crossroads: Defining the “why” before the “how”2.3.1 The offense play: Creating a new revenue stream2.3.2 The defense play: Boosting productivity and retention2.4 Measuring affect: The ROI boardroom test2.4.1 The ROI spectrum: From isolated pilots to enterprise-wide transformation2.4.2 Deconstructing value: Real-world use cases and their metrics2.5 Navigating the iceberg: Real-world obstacles and reality checks2.5.1 The unpredictable market2.5.2 The black box, hallucinations, and the ethical tightrope2.5.3 The human factor: Culture, talent, and augmentation2.6 The strategist’s bridge: From business “why” to technical “how”
3 Working with generative AI in finance
3.1 The new playbook: AI engineering3.1.1 The great shift: From custom models to foundation engines3.1.2 Finance’s reality: New engine, same high-stakes fuel3.1.3 Five truths for the modern financial AI engineer3.2 The GenAI blueprint: Architecting for finance3.2.1 The core building blocks of a financial AI system3.3 The builder’s journey: Powering Alpha Digest step-by-step3.3.1 Step 1—Choosing the foundation: The model selection dilemma3.3.2 Step 2—Grounding the model in reality: Mastering RAG3.3.3 Step 3—Teaching our style: The art of fine-tuning3.3.4 Step 4—The next frontier: From AI agents to agentic AI3.4 Deployment strategies and future directions3.4.1 Phased rollouts and continuous feedback3.4.2 Cost and performance optimization3.4.3 Governance, open source, and the future3.5 Wrapping up: From proofs of concept to scalable impact
Part 2 Building AI for credit risk
4 Risk and compliance: Credit risk fundamentals
4.1 From trust to scoring: How credit took shape4.1.1 The evolution from local judgments to standardized scoring4.1.2 The two pillars of repayment and the role of interest4.1.3 Modern channels: BNPL, micro-lending, and merchant cash advances4.2 Core domain concepts and the credit life cycle4.2.1 The loan life cycle: Personal and corporate4.2.2 Time-based performance and survival analysis4.2.3 Regional and product nuances4.3 A domain-driven four-layer framework4.3.1 Data asset layer4.3.2 Modeling layer4.3.3 Strategy and monitoring layer4.3.4 Application layer4.3.5 Why this matters and where we’re going next4.4 Essential metrics and techniques in credit scoring4.4.1 Revisiting core ML metrics: AUC, K-S, and PSI4.4.2 Binning and discretization: Fine classing vs. coarse classing4.4.3 Weight of Evidence and information value4.4.4 Handling special values, missing data, and outliers4.5 Why AI is indispensable and how it relates to building finance applications4.5.1 AI in strictly regulated, high-stakes lending4.5.2 Domain first, AI next: The best of both worlds4.5.3 The road to production: A preview of hands-on implementation4.5.4 The core takeaway
5 End-to-end credit scoring for financial applications: A real-world AI approach
5.1 Data pipeline: From raw data to credit-ready marts5.1.1 High-level BFSI data ingestion5.1.2 Constructing a credit data mart for lending5.1.3 A simple batch pipeline (Airflow)5.1.4 Feature stores: Evolving from batch pipelines to reusable assets5.2 Setting the stage: Realistic data and an end-to-end credit modeling flow5.2.1 Why tabular modeling dominates in BFSI5.2.2 Which datasets we’ll use and why5.2.3 Our end-to-end approach5.3 Implementing a basic credit model from scratch5.3.1 Starting our BFSI pipeline5.3.2 Data import and quick exploration5.3.3 WoE and IV calculation5.3.4 Executing WoE and IV, and then reviewing results5.3.5 XGBoost modeling with K-fold5.3.6 Visualizing ROC and a confusion matrix5.3.7 Credit score conversion5.3.8 Packaging and deploying the final BFSI model5.3.9 Beyond deployment: What “production-ready” really means
6 Enhancing BFSI scoring workflows: Advanced binning, monitoring, and explainability
6.1 Using OptBinning for automated binning and scorecards6.1.1 Automated binning with BinningProcess6.1.2 Building a scorecard (logistic model under the hood)6.1.3 Evaluating with ROC, CAP, and KS6.1.4 Monitoring model drift with ScorecardMonitoring6.2 Quick drift check with Evidently6.2.1 Creating the Evidently drift report6.2.2 Interpreting the Evidently dashboard6.3 Model interpretability: From black box to business dialogue6.3.1 Local explanations: Answering “Why me?” for the applicant6.3.2 Global explanations: Answering “How does this thing work?” for the auditor6.3.3 Putting explanations into practice6.3.4 Best practices for building trust6.4 Putting it all together: The four-layer framework in production6.5 The next frontier: GenAI and agentic workflows
Part 3 Building AI for fraud detection
7 Understanding financial fraud and foundational detection strategies
7.1 The criticality of fraud detection and prevention7.2 Demystifying financial fraud: Scope and typologies7.3 The distinctive nature of financial fraud7.4 Foundational approaches to fraud detection: The four-layer framework in action7.4.1 A glimpse into the framework: Foundational detection techniques7.4.2 Data asset layer for AI-powered fraud detection7.4.3 Modeling layer: AI techniques for fraud detection7.4.4 Strategy and monitoring layer: From scores to actionable intelligence7.4.5 Application layer: Deploying and operationalizing fraud defenses7.4.6 Synthesizing the layers: A blueprint for operational resilience
8 Building AI fraud detectors: From supervised scoring to anomaly identification
8.1 Recognizing the known: The power and nuances of supervised learning in fraud detection8.1.1 The cornerstone: Acquiring, managing, and wrestling with labeled data8.1.2 Beyond binary labels: Achieving contextual understanding in fraud classification8.1.3 Practical realities: The ongoing vigilance in supervised fraud modeling8.2 Data preparation and initial exploration for supervised modeling8.2.1 Loading and understanding our dataset: The Kaggle credit card fraud data8.2.2 Essential preprocessing: Scaling features and splitting the data8.3 Supervised fraud classification in action: Training and tuning a LightGBM model8.3.1 Defining the modeling pipeline and hyperparameter search space8.3.2 Training the model with randomized hyperparameter search8.3.3 Evaluating the tuned model: Comprehensive metrics on the test set8.3.4 Strategic insight: Analyzing performance at top-N% prediction thresholds8.4 Finding the unknown: Unsupervised anomaly detection with autoencoders8.4.1 Autoencoders: Learning normality to detect deviance8.4.2 Preparing data for the autoencoder8.4.3 Building and training the autoencoder model8.4.4 Calculating and examining reconstruction errors8.4.5 Visualizing reconstruction error distributions8.4.6 Setting practical thresholds in an unsupervised world8.4.7 Deploying real-time fraud scoring models: Practical considerations
9 Unveiling hidden connections: Graph-based analytics for advanced fraud detection
9.1 Why graph analytics are now essential in BFSI fraud detection9.2 Fundamentals of graph data in finance9.2.1 Core components: Nodes, edges, and properties9.2.2 Types of graphs in fraud detection and prevention9.2.3 The fundamental process for creating graph data for fraud analytics9.3 Hands-on: Building and analyzing a transaction graph with NetworkX9.3.1 Loading and initial inspection of the transaction dataset9.3.2 Constructing the multi-entity graph with NetworkX using parallel processing9.3.3 Community detection using the Louvain algorithm9.4 Toward advanced graph-based techniques: Graph Neural Networks9.4.1 Key design considerations for GNNs in fraud detection9.4.2 Hands-on: A simplified GNN for transaction fraud detection with PyTorch9.5 Synergistic fraud defense: Combining GNNs, LLMs, and future directions
Part 4 Building AI for investment strategies
10 Building the quantitative engine: Market direction prediction with machine learning
10.1 Architecting our AI-powered investment strategy10.1.1 Architecting our strategy: Aligning the playbook with the four-layer framework10.1.2 Defining our scope: A practical path to powerful results10.2 The data foundation: The lifeblood of a quant strategy10.2.1 The universe of financial data: A practitioner’s reality check10.2.2 Our strategic choice: Why ETFs as economic proxies?10.2.3 Building the ETF toolkit: Selecting economic barometers10.3 Acquiring and preparing the ETF data10.3.1 Beyond simple APIs: Industrial-strength data architecture10.4 Feature engineering: Crafting predictive signals10.4.1 Step 1: Defining the target variable (Y-label)10.4.2 Step 2: Engineering quant-grade predictive features10.4.3 Feature engineering realities10.4.4 From features to alpha: Next steps10.5 Predictive modeling: Selecting and validating robust ML methods10.5.1 The accuracy myth: Why high predictions demand scrutiny10.5.2 Our model choice: Why random forest excels for financial signals10.5.3 Time-series cross-validation: The foundation of intellectual honesty10.5.4 Building a robust validation scheme10.5.5 Training the final model with optimized parameters10.6 From signal to strategy: Next steps
11 Building the qualitative engine with news analysis and LLMs
11.1 The new playbook: From market numbers to market narratives11.1.1 The hard truth: A reality check on LLMs in investment research11.1.2 Our philosophy: The LLM as a probabilistic analyst, not a deterministic oracle11.1.3 The strategist’s hunt for alpha: Where qualitative signals shine11.1.4 Designing the qualitative engine architecture11.2 Implementing the news ingestion and processing pipeline11.2.1 The data sourcing reality: Building our own information infrastructure11.2.2 Our pragmatic toolkit: A four-source blend11.2.3 Implementing the fetcher functions11.3 Building the RAG feature pipeline: From raw news to retrieval-ready vectors11.3.1 Data unification and cleaning11.3.2 Chunking strategy: Choosing the right granularity11.3.3 Text embedding with OpenAI11.3.4 Vector database storage with ChromaDB11.3.5 Verification: Trust, but verify the pipeline11.4 Building the LLM Analyst: From retrieval to quantified signals11.4.1 Retrieval strategy: A two-step process for context and precision11.4.2 Structured prompt engineering: The art of instructing an analyst11.4.3 The final output: A structured, auditable signal
12 Augmenting investment decisions: LLM-guided portfolio allocation
12.1 The synthesis challenge: From separate engines to unified decisions12.1.1 The information asymmetry problem12.1.2 The synthesis workflow in our four-layer framework12.1.3 The role of macroeconomic indicators12.1.4 News sentiment vs. structured macro data12.2 Setting up the simulation infrastructure12.2.1 Defining the portfolio universe12.2.2 Timing assumptions12.3 Designing the system prompt12.3.1 Bounded discretion under constraints12.3.2 No date information12.3.3 Output structure12.4 Quantitative signal generation12.5 The LLM portfolio advisor12.6 Running the historical simulation12.6.1 What this simulation does and doesn’t show12.6.2 Path dependence, not learning12.7 Results and interpretation12.7.1 Interpreting the tradeoff12.7.2 Visualizing strategy behavior12.7.3 The decision log as audit trail12.8 Human oversight and the production gap12.8.1 Why automation isn't the goal12.8.2 The recommended workflow12.8.3 What this simulation omits12.8.4 Extensions and the path forward
Part 5 Building AI for customer and operational excellence
13 Enhancing customer experience: Building a personalization engine
13.1 The unique context of financial personalization13.1.1 From “add to cart” to “investment discovery”: The economics of attention13.1.2 Architecting the pipeline: From raw logs to a User 360 mart13.1.3 Understanding the AlphaStream dataset13.2 From rules to vectors: Modern user representation13.2.1 The engine of meaning: How to choose an embedding model13.2.2 Hands-on: Creating the Financial DNA vector with embeddings13.2.3 Discovering tribes: Using clustering to identify hidden user segments13.3 Matching intelligence: From user DNA to content discovery13.3.1 The content pipeline: From raw feed to vectors13.3.2 Hands-on: Real-time semantic matching13.3.3 Expanding horizons: Beyond simple vectors13.4 Strategy and governance: From algorithms to business impact13.4.1 Beyond clicks: Measuring financial success13.4.2 The fiduciary filter: Ethical guardrails13.4.3 From insights to organization-wide action
14 Driving operational efficiency: Building autonomous retention agents
14.1 The agent era: Extending the four-layer framework14.1.1 From rule-based automation to agentic reasoning14.1.2 The four-layer framework in the agent era14.1.3 The AlphaStream retention challenge14.2 Architecture: The retention intelligence pipeline14.2.1 The cost-aware design principle14.2.2 Pipeline components in detail14.2.3 Data flow and timing14.3 Hands-on: Building the retention copilot14.3.1 Loading data and filtering high-risk users14.3.2 Designing the tool interface14.3.3 Implementing the agent loop14.3.4 Testing the agent14.3.5 Batch processing multiple users14.3.6 Aggregating results by cluster14.4 Connecting to operations: Slack integration14.5 Human-in-the-loop and governance14.5.1 The Draft-Approve pattern14.5.2 Governance policies and guardrails14.6 Measuring agent impact14.6.1 Defining success metrics14.6.2 Experimental design for agent evaluation14.6.3 Return on investment calculation
appendix A Setting up your development environment
A.1 PrerequisitesA.2 Creating a virtual environmentA.2.1 Using venv (recommended for most users)A.2.2 Using conda (alternative)A.3 Installing core librariesA.4 Managing environment variablesA.5 Cloning the book’s GitHub repositoryA.6 Verifying the installation
appendix B Comprehensive fraud typology and modeling reference
B.1 Financial fraud typologiesB.2 Feature engineering cheat sheet for fraudB.3 Algorithm tradeoff matrixB.4 Designing a layered fraud defense strategy
appendix C Dataset and feature dictionaries
C.1 Credit risk dataset (part 2)C.2 Fraud detection dataset (part 3)C.3 Market intelligence and news datasets (part 4)C.4 Synthetic retail banking and interaction dataset (part 5)
appendix D A quick reference on key financial AI metrics
D.1 Credit risk and stability metrics (part 2)D.2 Fraud and imbalanced classification metrics (part 3)D.3 Investment and market prediction metrics (part 4)D.4 LLM, RAG, and agentic workflow metrics (part 5)
appendix E Translating regulation into engineering requirements
E.1 Model risk and validation (chapters 4–9)E.2 Fair and explainable credit decisions (chapters 4–6)E.3 AI-specific regulation (the fast-moving frontier)E.4 Communications, data protection, and automated decisions (parts 4–5)E.5 AML, payments, and operational resilience (part 3 and cross-cutting)E.6 The pre-production checklist
appendix F References and further reading
F.1 Chapter 1: Introducing AI in financeF.1.1 ReferencesF.1.2 Further readingF.2 Chapter 2: From model to money: A strategist’s guide to real-world financial AIF.2.1 ReferencesF.2.2 Further readingF.3 Chapter 3: Working with Generative AI in financeF.3.1 ReferencesF.3.2 Further readingF.4 Chapter 4: Risk and compliance: Credit risk fundamentalsF.4.1 ReferencesF.4.2 Further readingF.5 Chapter 5: End-to-end credit scoring for financial applications: A real-world AI approachF.5.1 ReferencesF.5.2 Further readingF.6 Chapter 6: Enhancing BFSI scoring workflows: Advanced binning, monitoring, and explainabilityF.6.1 ReferencesF.6.2 Further readingF.7 Chapter 7: Understanding financial fraud and foundational detection strategiesF.7.1 ReferencesF.7.2 Further readingF.8 Chapter 8: Building AI fraud detectors: From supervised scoring to anomaly identificationF.8.1 ReferencesF.8.2 Further readingF.9 Chapter 9: Unveiling hidden connections: Graph-based analytics for advanced fraud detectionF.9.1 ReferencesF.9.2 Further readingF.10 Chapter 10: Building the quantitative engine: Market direction prediction with machine learningF.10.1 ReferencesF.10.2 Further readingF.11 Chapter 11: Building the qualitative engine with news analysis and LLMsF.11.1 ReferencesF.11.2 Further readingF.12 Chapter 12: Augmenting investment decisions: LLM-guided portfolio allocationF.12.1 ReferencesF.12.2 Further readingF.13 Chapter 13: Enhancing customer experience: Building a personalization engineF.13.1 ReferencesF.13.2 Further readingF.14 Chapter 14: Driving operational efficiency: Building autonomous retention agentsF.14.1 ReferencesF.14.2 Further reading

Content preview from Financial AI in Practice

12 Augmenting investment decisions: LLM-guided portfolio allocation

This chapter covers

Integrating quantitative momentum signals
Designing prompts with bounded discretion
Simulating with point-in-time information
Comparing simulations with live trading
Keeping human judgment in final decisions

In chapter 10, we built a quantitative engine that processes the market’s numerical signals: price movements, momentum, and cross-asset correlations. In chapter 11, we constructed a qualitative engine that extracts signals from text: news sentiment, policy tone, and emerging risks. Each engine captures different information. The quantitative engine detects statistical patterns but can’t explain the drivers behind price movements. The qualitative ...

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Publisher Resources

ISBN: 9781633435391Publisher Support Publisher Website Purchase Link

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills

Financial AI in Practice

by Taehun Kim