GraphRAG, ColBERT & LangGraph Labs

Neo4j GraphRAG credit underwriting, ColBERT retrieval, LangGraph agents, PEFT/distillation, and classic ML playgrounds.

Run live production ML labs: Neo4j GraphRAG credit agents, ColBERT code search, LangGraph ad loops, PEFT distillation, multipath voting, and fail-closed gates.


LLM FINE-TUNING

LLM Fine-Tuning

PEFT, distillation, and multi-adapter serving playgrounds.

Edge Distillation with Multi-Stage Tuning Pipeline for SLMs

Engineer a high-fidelity SLM for interactive persona by distilling linguistic patterns from frontier models (GPT 5.4 mini).

  • Distill latent reasoning and Chain-of-Thought (CoT) capabilities from GPT-5.4 into a 3B model.

  • Engineer multi-stage tuning pipeline - SFT for grounding, RKD for logic, and DPO for stylistic parity.

  • Standardize input/output schemas using chat templates.

  • Implement 4-bit quantization (GGUF) to balance VRAM efficiency and perplexity for edge hardware.

  • Deploy via AWS SageMaker LMI/vLLM engine for paged-attention concurrency and real-time streaming.

Explore
Edge Distillation with Multi-Stage Tuning Pipeline for SLMs — LLM Fine-Tuning lab by Kuriko IWAI

Advanced PEFT - LoRA Multi-Adapter Orchestration

Orchestrate multiple LoRA adapters for Dialect Reconstruction, PII Masking, and Sentiment Neutralization. Optimize PEFT workflows on AWS SageMaker.

  • Dynamic switching between specialized LoRA weights without model reloading.

  • Native adapters for Dialect Reconstruction, PII Redaction, Sentiment Neutralization, and Style Transfer.

  • Granular hyperparameter tuning via real-time adjustment of Rank, Alpha, and Target Modules (Attention vs. FF).

  • Compute-efficient inference optimized for ml.g5.xlarge instances with VRAM telemetry monitoring.

  • AWS SageMaker integration for Serverless or Multi-Model Endpoints (MME).

  • Live trace observability via debug adapter retrieval latency and S3 artifact injection logs in real-time.

Explore
Advanced PEFT - LoRA Multi-Adapter Orchestration — LLM Fine-Tuning lab by Kuriko IWAI

AGENT SYSTEMS

Agent Systems

Multi-agent orchestration, GraphRAG decision agents, ColBERT retrieval, and LangGraph loops.

Orchestrating Autonomous Agent Networks

A Python framework for building autonomous agent networks with multi-step reasoning. Automated formation, TaskGraph orchestration, and model-agnostic optimization.

  • Autonomous Agent Formations (Solo, Supervising, Squad, Random)

  • Graph-Based Task Execution via TaskGraph (Nodes & Edges)

  • Model-Agnostic LLM Curation with LiteLLM integration

  • Advanced Memory Management using mem0ai and Chroma DB

  • Built-in RAG Support and External Tooling via Composio

  • Automated Workflow Optimization and Dependency Resolution

Explore
Orchestrating Autonomous Agent Networks — Agent Systems lab by Kuriko IWAI

Credit Underwriting Agent: GraphRAG, Self-Consistency & Release Gates

Build a Neo4j GraphRAG credit agent: hop-decayed related-party risk, RRF fusion, multipath voting, AI gateway, and fail-closed release gates on Databricks Apps.

  • Hybrid retrieval with Reciprocal Rank Fusion (RRF): SQL features + Neo4j ownership hops + BM25/vector policy context.

  • Related-party GraphRAG: hop-decayed group exposure surfaces submerged ~$12M NPL / deficit under shell subsidiaries.

  • Signature fork: related-party commercial parent (multipath → REVIEW) vs clean micro / simple-retail (light path + gates).

  • Multi-path self-consistency: five underwriting personas, critic-weighted votes, circuit breaker → HITL REVIEW.

  • AI control plane: ingress ACL/rate limits, SQL/tool allow-lists, egress sanitization, red-team + fairness checks.

  • Release engineering: offline eval gates, canary evidence packets, fail-closed APPROVE path on Databricks Apps.

Explore
Credit Underwriting Agent: GraphRAG, Self-Consistency & Release Gates — Agent Systems lab by Kuriko IWAI

CodeContext: Low-Latency Neural Code Search with ColBERT

Try ColBERT late-interaction code search live: AST chunking, Redis cache hits, token heatmaps, and sub-100ms latency on a FastAPI retrieval stack.

  • Interactive ColBERT late-interaction search over indexed code snippets.

  • Token heatmap visualization showing which tokens drive MaxSim alignment.

  • Redis semantic cache path with HIT / MISS latency telemetry.

  • AST / Tree-sitter structural chunking for function-level retrieval.

  • FastAPI microservice pattern ready for App Runner + ElastiCache.

  • RAGAS-ready evaluation hooks and OpenLLMetry span tracing.

Explore
CodeContext: Low-Latency Neural Code Search with ColBERT — Agent Systems lab by Kuriko IWAI

Shipping AI Systems?

I help teams design and deploy scalable ML / RAG / LLM pipelines and MLOps infrastructure.



Or explore:

Ad-Creative Agent: Self-Optimizing Creatives with LangGraph

Run a live LangGraph Generator→Critic loop with synthetic CTR rewards, prompt refinement, and Bayesian A/B winner selection for ad creatives.

  • Interactive LangGraph Generator → Critic → Prompt Refiner optimization loop.

  • Synthetic CTR reward signals scored against historical angle priors.

  • Bayesian A/B (Thompson / beta-binomial) winner selection before launch.

  • Workflow trace showing inference-time System-2 self-correction steps.

  • Campaign brief controls: product, audience, offer, channel, iterations.

  • FastAPI + Redis caching pattern ready for App Runner + ElastiCache.

Explore
Ad-Creative Agent: Self-Optimizing Creatives with LangGraph — Agent Systems lab by Kuriko IWAI

CLASSIC MACHINE LEARNING

Classic Machine Learning

Demand modeling & MLOps, SVD/PCA, and interactive regression / loss landscapes.

Bayesian Demand Modeling & Production MLOps

Scale retail revenue with Bayesian-optimized price elasticity. Features a multi-model ensemble (DLN, LightGBM, SVR), DVC lineage, and AWS Lambda deployment.

  • Multi-model failover system (PyTorch deep learning model, LightGBM, SVR, Elastic Net).

  • HPO via Bayesian optimization with Optuna.

  • Low-latency caching with ElastiCache Redis.

  • Weekly-scheduled ML lineage management with DVC & Prefect.

  • Automated data drift and fairness/bias testing (SHAP).

  • CI/CD integration with GitHub Actions, AWS CodeBuild, and Snyk for security scanning.

Explore
Bayesian Demand Modeling & Production MLOps — Classic Machine Learning lab by Kuriko IWAI

SVD Image Compression & PCA Deep Dive

Apply Singular Value Decomposition for Principal Component Analysis. Interactive image compression. Incremental, Randomized, and Kernel PCA.

  • Interactive SVD Rank Adjustment

  • Step-by-step Mathematical Derivation of PCA

  • Comparison of 5 PCA methods (Incremental, Kernel, etc.)

  • Real-world Telecom Churn Data Simulation

  • Low-Rank Approximation Visualizations

Explore
SVD Image Compression & PCA Deep Dive — Classic Machine Learning lab by Kuriko IWAI

Visualizing Regression: From Loss Functions to Generalization Bounds

Explore MSE, MAE, L1/L2 regularization, generalization bounds with interactive explorers for loss functions and model complexity.

  • Interactive Loss Function Explorer (MSE, MAE)

  • Regularization Strength (λ) Simulator

  • Comparison of Parametric vs Non-parametric models

  • Generalization Bound Mathematical Analysis

  • Real-world Regression Scenarios & MSE Results

Explore
Visualizing Regression: From Loss Functions to Generalization Bounds — Classic Machine Learning lab by Kuriko IWAI

Shipping AI Systems?

I help teams design and deploy scalable ML / RAG / LLM pipelines and MLOps infrastructure.



Or explore: