Front Matter
7 entries- F1ForewordDiscovery as the common thread across software engineering, scientific research, and autonomous AI.
- F2What This Book CoversThe seven-part arc: foundations, vibe coding, data and models, knowledge, simulation, domains, and autonomous systems.
- F3Who Should Read This BookThree personas: the software engineer building research tools, the scientist who wants rigorous theory and working code, and the domain expert adding AI to a scientific workflow.
- F4How to Use This BookReading paths for graduate research, software practitioner, and domain scientist audiences; how the parts depend on each other.
- F5About the AuthorsAlexander (Sasha) Apartsin, Ph.D. and Yehudit Aperstein, Ph.D.
- F6Copyright & LegalEdition, license, and attribution.
- F7About the Hands-On AI Science SeriesThe nine-book Hands-On AI Science series and where this volume fits.
Part I · Foundations of Discovery AI
6 chaptersDiscovery as search, inference, optimization, and experimentation. The scientific, mathematical, and architectural foundations all later chapters build on.
-
1Discovery as Search Search spaces, objective functions, utility, regret, information gain, exploration versus exploitation, and the role of AI in accelerating search.
part-1-foundations/module-01-discovery-as-search/ -
2Scientific Discovery and Knowledge Creation Scientific method, falsification, reproducibility, Bayesian science, measurement, uncertainty, and construct validity.
part-1-foundations/module-02-scientific-discovery/ -
3Knowledge Representation Logic, ontologies, knowledge graphs, embeddings, vector databases, and hybrid retrieval.
part-1-foundations/module-03-knowledge-representation/ -
4Reasoning for Discovery Deductive, inductive, abductive, analogical, and causal reasoning; chain-of-thought in language models; AlphaGeometry and formal mathematical reasoning.
part-1-foundations/module-04-reasoning-for-discovery/ -
5Discovery Through Data, Models, and Simulation Data-driven, theory-driven, and simulation-driven discovery; surrogate models, digital twins, and active learning.
part-1-foundations/module-05-data-models-simulation/ -
6Discovery System Architecture The Discovery Workbench: artifact graphs, provenance, human approvals, sandboxes, cost, and observability.
part-1-foundations/module-06-system-architecture/
Part II · Software Engineering & Vibe Coding
18 chaptersVibe coding as a rigorous discipline: specification, steering, verification, and repair. From a first coding agent to fully autonomous software organizations.
-
7Software Development as a Discovery Process SDLC as iterative hypothesis testing; agile, lean, and design thinking as discovery methodologies.
part-2-software-engineering/module-07-software-discovery/ -
8Foundations of AI Assisted Software Engineering Code foundation models, program synthesis, repository-level reasoning, and human-AI collaboration patterns.
part-2-software-engineering/module-08-ai-software-foundations/ -
9Vibe Coding as Specification, Steering, Verification, and Repair Natural language specification, intent elicitation, steering by tests and artifacts, verification, and the formalized vibe coding loop.
part-2-software-engineering/module-09-vibe-coding/ -
10Prompting to Programming Structured outputs, tool definitions, typed agent interfaces, DSPy optimization, and reusable agent components.
part-2-software-engineering/module-10-prompting-to-programming/ -
11Context Engineering at Repository Scale Repository graphs, chunking strategies, retrieval-augmented code generation, context compression, and agent memory.
part-2-software-engineering/module-11-context-engineering/ -
12Building MCP Servers for Scientific Workflows Model Context Protocol architecture, tool and resource definitions, scientific tool integration, testing, and publishing.
part-2-software-engineering/module-12-mcp-servers/ -
13Discovery of Requirements Stakeholder goals, AI-assisted requirement extraction, validation, traceability, and conflict detection.
part-2-software-engineering/module-13-requirements-discovery/ -
14Discovery of Architectures Architectural styles, design patterns, AI-assisted architecture generation, tradeoff analysis, and fitness functions.
part-2-software-engineering/module-14-architecture-discovery/ -
15Discovery of Algorithms Algorithm search, complexity analysis, program synthesis, and benchmarking methodology.
part-2-software-engineering/module-15-algorithm-discovery/ -
16AI Assisted Implementation at Repository Scale Multi-file code generation, impact analysis, change propagation, patch review, and failure patterns.
part-2-software-engineering/module-16-ai-implementation/ -
17Multi-Agent Software Teams Agent roles, workflow graphs, shared state, debate and critique patterns, human approval gates, and failure recovery.
part-2-software-engineering/module-17-multi-agent-teams/ -
18AI Assisted Testing and QA Test generation, property-based testing, invariant discovery, mutation testing, and evaluation of AI-generated tests.
part-2-software-engineering/module-18-ai-testing/ -
19AI Assisted Debugging Hypothesis-driven debugging, root cause analysis, trace-driven reproduction, self-debugging agents, and error budgets.
part-2-software-engineering/module-19-ai-debugging/ -
20AI for Software Security Threat modeling, secure SDLC, vulnerability discovery, AI-assisted security review, and sandboxing coding agents.
part-2-software-engineering/module-20-software-security/ -
21AI for DevOps and Platform Engineering CI/CD pipelines, observability, infrastructure as code, incident analysis, and deployment automation.
part-2-software-engineering/module-21-devops/ -
22MLOps, LLMOps, and AgentOps Data pipelines, model lifecycle, prompt versioning, agent trace logging, governance, and rollback.
part-2-software-engineering/module-22-mlops-llmops/ -
23Evaluating AI Coding Agents SWE-bench tasks, repository task suites, human review, benchmark validity, cost, latency, and reliability.
part-2-software-engineering/module-23-evaluating-agents/ -
24Autonomous Software Organizations SWE agents, autonomous software factories, issue triage, review and release pipelines, and self-improving systems.
part-2-software-engineering/module-24-autonomous-software/
Part III · Data and Models
11 chaptersScientific foundation models, multimodal AI, reasoning models, and generative models for molecular design. The machine learning toolkit of the AI scientist.
-
25Exploratory Discovery Exploratory data analysis, clustering, dimensionality reduction, pattern discovery, and hypothesis generation from data.
part-3-data-models/module-25-exploratory-discovery/ -
26Representation Learning Self-supervised and contrastive learning, latent geometry, foundation model embeddings, and probing evaluation.
part-3-data-models/module-26-representation-learning/ -
27Scientific Foundation Models Protein, molecular, materials, and genomics foundation models; ESM-3, AlphaFold3, GNoME, Uni-Mol2; fine-tuning with LoRA; scaling laws for scientific domains.
part-3-data-models/module-27-foundation-models/ -
28Multimodal Scientific AI Molecule-text, protein-text, and vision-language models; cross-modal retrieval; PaperQA2; multimodal hypothesis generation.
part-3-data-models/module-28-multimodal-scientific-ai/ -
29Reasoning Models for Discovery Chain-of-thought, test-time compute, process reward models, MCTS over reasoning, AlphaProof, FunSearch, and formal theorem proving with Lean4.
part-3-data-models/module-29-reasoning-models/ -
30Anomaly and Novelty Discovery Density-based, isolation-based, and deep anomaly detection; open-world learning; drift detection; separating errors from discoveries.
part-3-data-models/module-30-anomaly-novelty/ -
31Causal Discovery and Causal Inference Structural causal models, causal discovery algorithms, treatment effects, do-calculus, counterfactuals, and double ML.
part-3-data-models/module-31-causal-discovery/ -
32Bayesian Discovery and Uncertainty Bayesian workflow, MCMC, variational inference, posterior predictive checking, model comparison, and experimental design.
part-3-data-models/module-32-bayesian-discovery/ -
33Scientific Machine Learning PINNs, neural operators, equivariant networks, universal force fields, hybrid models, and differentiable simulation.
part-3-data-models/module-33-scientific-ml/ -
34Generative Models for Discovery VAEs, diffusion, flow matching, equivariant architectures, graph generation, protein-ligand cofolding, and property-guided sampling.
part-3-data-models/module-34-generative-models/ -
35Symbolic Regression and Equation Discovery Genetic programming, neural-guided symbolic regression, dimensional constraints, interpretability, and parsimony.
part-3-data-models/module-35-symbolic-regression/
Part IV · Knowledge
6 chaptersLiterature at scale, retrieval-augmented discovery, knowledge graphs, hypothesis generation, and research agents. The full knowledge infrastructure for AI-assisted science.
-
36Literature Mining Citation analysis, semantic search, knowledge extraction, claim and hypothesis extraction, and domain mapping.
part-4-knowledge/module-36-literature-mining/ -
37Retrieval Augmented Discovery Systems RAG architecture, research copilots, evidence grounding, PaperQA2, and citation verification.
part-4-knowledge/module-37-rag-discovery/ -
38Knowledge Graph Discovery Link prediction, graph embeddings, GNNs for knowledge graph completion, ontology enrichment, and relation extraction.
part-4-knowledge/module-38-knowledge-graphs/ -
39Hypothesis Generation Gap analysis, analogical transfer, AI-generated hypotheses, scoring by plausibility, novelty, and testability.
part-4-knowledge/module-39-hypothesis-generation/ -
40Research Agents Literature, coding, analysis, and reviewer agents; Coscientist, ChemCrow, AI Co-Scientist; memory and long-horizon planning.
part-4-knowledge/module-40-research-agents/ -
41Scientific Claim Validation Claim extraction, evidence mapping, citation checking, artifact verification, and reproducibility auditing.
part-4-knowledge/module-41-claim-validation/
Part V · Simulation and Optimization
6 chaptersDifferentiable programming, scientific simulation, world models, Bayesian optimization, automated experiment design, and provenance. The computational machinery of discovery.
-
42Differentiable Programming for Discovery Automatic differentiation, JAX transforms, differentiable physics, adjoint methods, and differentiable molecular dynamics.
part-5-simulation-optimization/module-42-differentiable-programming/ -
43Scientific Simulation Agent-based models, Gillespie algorithm, molecular dynamics, synthetic data, and approximate Bayesian computation.
part-5-simulation-optimization/module-43-scientific-simulation/ -
44World Models for Discovery Recurrent and transformer world models, counterfactual reasoning, model-based planning, and uncertainty estimation.
part-5-simulation-optimization/module-44-world-models/ -
45Optimization for Discovery Bayesian optimization, evolutionary algorithms, reinforcement learning, multi-objective Pareto frontiers, and expensive function optimization.
part-5-simulation-optimization/module-45-optimization/ -
46Automated Experiment Design Active learning, sequential design, bandit algorithms, adaptive trials, closed-loop optimization, and stopping rules.
part-5-simulation-optimization/module-46-experiment-design/ -
47Experiment Registries and Scientific Provenance Experiment registry design, artifact versioning, data lineage, configuration management, and reproducible report generation.
part-5-simulation-optimization/module-47-experiment-registries/
Part VI · Scientific Domains
5 chaptersDeep application chapters: the AI toolkit deployed in biology, chemistry, physics, climate science, and social systems. Domain models, benchmarks, and end-to-end recipes.
-
48Discovery AI for Biology and Medicine Drug discovery, AlphaFold, de novo protein design, single-cell genomics, clinical AI, and validation constraints.
part-6-scientific-domains/module-48-biology-medicine/ -
49Discovery AI for Chemistry and Materials De novo molecular generation, protein-ligand cofolding, reaction prediction, crystal design, universal force fields, and autonomous labs.
part-6-scientific-domains/module-49-chemistry-materials/ -
50Discovery AI for Physics and Engineering Symbolic regression for physics laws, PINNs, neural operators, engineering optimization, and simulation-based inference.
part-6-scientific-domains/module-50-physics-engineering/ -
51Discovery AI for Climate and Earth Science AI weather prediction, climate emulation, extreme event detection, carbon accounting, and uncertainty quantification.
part-6-scientific-domains/module-51-climate-earth/ -
52Discovery AI for Social and Economic Systems Market discovery, network analysis, behavioral modeling, causal inference in social data, and policy simulation.
part-6-scientific-domains/module-52-social-economic/
Part VII · Autonomous Discovery
6 chaptersAI scientists, multi-agent discovery teams, self-driving laboratories, system evaluation, governance, and the research frontier of autonomous science.
-
53AI Scientists Automated research loops; AI Scientist v1 and v2, Coscientist, AI Co-Scientist, FunSearch; human supervision and research integrity.
part-7-autonomous-discovery/module-53-ai-scientists/ -
54Multi-Agent Discovery Systems Scientific agent teams, debate and adversarial critique, peer review simulation, collective intelligence, and coordination mechanisms.
part-7-autonomous-discovery/module-54-multi-agent-discovery/ -
55Self-Driving Laboratories Laboratory robotics, closed-loop experimentation, ChemOS, digital-physical interfaces, and safety constraints.
part-7-autonomous-discovery/module-55-self-driving-labs/ -
56Evaluating Discovery Systems ScienceAgentBench, DiscoveryBench, MLE-bench, SWE-bench, GPQA Diamond, FrontierMath; construct validity and benchmark saturation.
part-7-autonomous-discovery/module-56-evaluating-discovery/ -
57Responsible Discovery AI Safety, dual-use risks, scientific integrity, alignment, governance, publication ethics, and regulatory frameworks.
part-7-autonomous-discovery/module-57-responsible-discovery/ -
58Future Directions Autonomous innovation, human-AI co-discovery, AI research organizations, open problems, and scientific institutions in the agent era.
part-7-autonomous-discovery/module-58-future-directions/
Appendices
7 appendices- AMathematical FoundationsProbability, statistics, information theory, optimization, linear algebra, graph theory, Bayesian inference, causal inference, and experimental design.
- BPython for Discovery AIScientific Python, reproducible notebooks, command-line tools, packaging, testing, type checking, and data pipelines.
- CVibe Coding Practitioner GuideSpecifying intent, steering by tests, inspecting diffs, recovering from bad changes, context engineering, and red-teaming prompts.
- DDiscovery AI Tool EcosystemCoding agents, agent frameworks, scientific foundation model registry, retrieval systems, optimization libraries, and generative model zoo.
- EResearch Paper Reading GuideHow to read a paper with AI assistance: extracting claims, checking assumptions, mapping evidence, and using PaperQA2.
- FReproducibility and Artifact GuideData versioning, environment capture, experiment registries, result tables, claim-to-artifact traceability, and reproducibility packages.
- GDiscovery AI Capstone ProjectsSix capstone tracks: graduate research, software practitioner, scientific computing, knowledge systems, generative discovery, and multimodal science.