Skip to content
K
KGAIT Solutions
Portfolio

Our Work

Here are some of the projects KGA has delivered. You can view a demo for each project.

AI / Machine Learningベータ(Beta)
Since Mar 2026·3 on team

AIRA — AI Chatbot Platform

A research SaaS automating customer support. Uses natural language processing to build a 24/7 customer response system.

0.88検索 nDCG@10+42%
RAGVector SearchLLM Gateway+3
View Demo
Automation / Logistics開発中(In development)
Since Mar 2026·2 on team

HAYAKU — Delivery Management System

A delivery management platform unifying real-time tracking, route optimization, and driver management.

22%総走行距離削減vs 貪欲法
VRPReal-timeGeospatial+3
View Demo
Automation / E-Commerceベータ(Beta)
Since Mar 2026·2 on team

KANRI — Inventory Automation Tool

A cloud tool centralizing inventory across multiple e-commerce storefronts and automating reorders.

12s同期ラグ中央値
Event SourcingCQRSIdempotency+3
View Demo
SaaS / F&Bベータ(Beta)
Since Mar 2026·2 on team

TENJIN — Restaurant POS System

A tablet-based cloud POS system integrating order management, kitchen display, and sales analytics.

280ms同期ラグ中央値
Offline-firstCRDTEdge Compute+3
View Demo
SaaS / Real Estate研究開発(Research)
Since Apr 2026·2 on team

SUMIKA — Real Estate Platform

A real-estate search and management platform with VR tours. End-to-end from listing to customer CRM.

9MBシーンサイズ-92%
3DWebGPUGaussian Splatting+3
View Demo
MicroSaaS / HRベータ(Beta)
Since Mar 2026·1 on team

JINZAI — HR Recruitment MicroSaaS

A lean recruiting tool for SMBs. Applicant tracking, interview scheduling, and evaluation management without the bloat.

91%評価観点一致率
HRTechATSDEI+3
View Demo
IoT / Manufacturing研究開発(Research)
Since Mar 2026·2 on team

KOJO — IoT Factory Monitoring

An IoT dashboard for real-time production line monitoring. Anomaly detection, predictive maintenance, and efficiency analytics unified.

18ms推論レイテンシ
Edge AIIoTTinyML+3
View Demo
Analytics / E-Commerce本番稼働(Production)
Since Mar 2026·2 on team

INSIGHT — E-Commerce Analytics Dashboard

A real-time analytics platform for e-commerce sales, customer behavior, and marketing effectiveness.

850ms取込→可視化 P95
AnalyticsStreamingOLAP+3
View Demo
SaaS / Healthcare研究開発(Research)
Since Apr 2026·2 on team

YOYAKU — Medical Appointment System

An integrated online booking, intake, and chart-management system for clinics and hospitals. Drastically improved patient experience.

92%FHIRカバレッジ
HealthcareFHIRLINE+3
View Demo
Fintech / Payments研究開発(Research)
Since Mar 2026·3 on team

ZENITH — Fintech Payment Dashboard

A unified management dashboard for multiple payment gateways. Fraud detection, revenue forecasting, and automated reporting built in.

+4.1pt認可成功率改善
PaymentsOrchestrationPCI DSS+3
View Demo
MicroSaaS / Productベータ(Beta)
Since Mar 2026·2 on team

NAVI — SaaS Onboarding Tool

A no-code onboarding builder for SaaS products. Product tours, checklists, and NPS surveys unified.

0.9KBコアSDK
SDKGrowthWeb Components+3
View Demo
Security / SaaS開発中(In development)
Since Apr 2026·3 on team

MIRU — Security Monitoring Platform

Real-time enterprise network security monitoring. Threat detection, incident response, and compliance reporting automated.

180標準ルール数
SecurityeBPFSIEM+3
View Demo
AI / Machine Learning開発中(In development)
Since Apr 2026·2 on team

Kotoba — Japanese RAG Evaluation Harness

A reproducible RAG evaluation harness designed for Japanese-language documents. Combines Sudachi/MeCab morphological pre-processing, pgvector-based batch measurement of nDCG@10, MRR, and Recall@K, and automated comparisons across Japanese-tuned embeddings such as ruri-large. Internal R&D harness.

1,200 QA評価データセット件数+1,200 (内部合成)
RAG評価日本語NLP+2
View Demo
AI / Machine Learning研究開発(Research)
Since Apr 2026·2 on team

OpenSafe — OSS Agent Evaluation Sandbox

An OSS sandbox that isolates LLM-agent tool execution under least privilege. Combines a LangGraph-based planner, OPA/Rego capability gates, gVisor isolation, and OpenTelemetry gen_ai semantic-convention tracing to block unsafe tool calls before they fire. Internal R&D prototype.

99.0%悪性プロンプト遮断率198/200
エージェントセキュリティOPA+2
View Demo
AI / Machine Learning開発中(In development)
Since May 2026·2 on team

LoRA-JP — Japanese Domain LoRA Fine-tuning Pipeline

A reproducible QLoRA/LoRA fine-tuning pipeline for Japanese business documents. Uses Unsloth+TRL to accelerate training and vLLM dynamic LoRA adapters for hot-swappable serving. Internal R&D project.

1.9x学習スループットvs HF標準
LoRAファインチューニングvLLM+2
View Demo
AI / Machine Learningベータ(Beta)
Since Apr 2026·3 on team

Kagami — Multi-provider LLM Gateway

An internal LLM gateway that fronts multiple providers behind one unified API, with PII redaction, cost-aware routing, rate limiting, and audit logging. Built on a Go control plane, a Rust data plane, Envoy ext_authz, and Microsoft Presidio with Japanese-language recognizers.

96.5%PII検出率(社内テスト)内部評価
LLMゲートウェイPIIコスト最適化+2
View Demo
AI / Machine Learning開発中(In development)
Since May 2026·2 on team

Aokigahara — Continuous LLM Evaluation Pipeline

An EvalOps pipeline that runs regression suites against every prompt, model, or adapter change. Integrates Ragas, DeepEval, and Promptfoo, schedules runs on Argo Workflows, and persists results to DuckDB. Internal R&D prototype.

14分PR評価完了時間社内運用
EvalOpsCI/CDRagas+2
View Demo
AI / Machine Learning研究開発(Research)
Since Apr 2026·2 on team

Vision Hokusai — Self-supervised Vision Backbone for Edge Inference

An edge-focused vision backbone learned via DINOv3-style self-supervision, distilled into MobileViTv3, and pushed through INT8 quantization on OpenVINO/CoreML. Targets near-zero-shot transfer to classification, detection, and segmentation. Internal R&D.

81.4%教師モデル Linear ProbeImageNet-1k
自己教師あり学習エッジAI蒸留+2
View Demo
AI / Machine Learningベータ(Beta)
Since Apr 2026·2 on team

Koe Kaibou — Real-time JA Speech Diarization and Translation

A real-time pipeline that diarizes meeting audio per speaker while simultaneously translating Japanese into English, Portuguese, and Chinese. Built on pyannote.audio with LoRA adaptation, INT8-quantized whisper.cpp, and NLLB-200-3.3B served via vLLM 0.7.

1.4sE2Eレイテンシp95
音声認識話者分離機械翻訳+2
View Demo
AI / Machine Learning開発中(In development)
Since Apr 2026·3 on team

Mirai Multimodal — Multimodal RAG Platform

A multimodal RAG pipeline that indexes PDFs, audio, and video into a unified vector space and returns cross-modal search and generative answers. PDFs run through TrOCR for figure-text extraction, audio through Whisper-v3, and video through joint frame+audio embedding, with pgvector and LanceDB on the storage side and NVIDIA NIM on the inference side.

0.88PDF抽出 F1TrOCR単体 0.71
マルチモーダルRAGベクトル検索+2
View Demo
AI / Machine Learning研究開発(Research)
Since Apr 2026·2 on team

Synth Umi — Synthetic Data Generation with DP-SGD

A research prototype that generates synthetic data for both tabular and text modalities under differential privacy (DP-SGD). Tabular uses DP-CTGAN; text combines Llama 3.2 1B with LoRA + DP-SGD; Opacus handles ε/δ accounting throughout.

0.84表データF1 (ε=3)実データ 0.91
差分プライバシー合成データLoRA+2
View Demo
AI / Machine Learningベータ(Beta)
Since Apr 2026·3 on team

Guard Tate — LLM Output Safety Layer

A drop-in safety proxy for existing LLM API calls that detects and blocks prompt injection, PII leakage, and unsafe outputs. Combines a DeBERTa-v3 classifier, Microsoft Presidio, and an OPA policy engine behind an OpenAI-compatible API.

0.93インジェクション F1HarmBench subset
LLMセキュリティプロンプトインジェクションPII+2
View Demo
Quantum Computing研究開発(Research)
Since Apr 2026·2 on team

QSIM-YAMATO — GPU-Accelerated Quantum Circuit Simulator

An internal R&D prototype of a GPU-accelerated quantum circuit simulator built on NVIDIA cuQuantum (cuStateVec / cuTensorNet). Used to evaluate variational quantum algorithms (VQE / QAOA) at ≤32 qubits via classical simulation for NISQ-era design exploration. This is classical simulation — not real quantum hardware.

32 qubit対象 qubit 上限 (state vector)
quantumsimulationgpu+3
View Demo
Quantum Computing研究開発(Research)
Since Apr 2026·2 on team

PQC-MUSASHI — Post-Quantum Cryptography Migration Toolkit

An internal toolkit for staged migration to post-quantum cryptography centered on the NIST-standardized ML-KEM (FIPS 203), ML-DSA (FIPS 204), and Falcon. Built on liboqs and OpenSSL 3.3 oqs-provider, it covers hybrid TLS, certificate inventory, and dependency scanning.

FIPS 203 / 204 / 205対象標準
quantumpqccryptography+3
View Demo
Quantum Computing研究開発(Research)
Since May 2026·2 on team

QML-SAKURA — Quantum-Classical Hybrid Machine Learning

Internal R&D bridging Pennylane variational quantum classifiers (VQC) with PyTorch / JAX classical networks. Scoped to small datasets, the focus is empirical study of barren-plateau detection and parameter-initialization strategies. All results are from classical simulation, not quantum hardware.

≤2k サンプル (研究用途)対象データ規模
quantumqmlmachine-learning+2
View Demo
Quantum Computing研究開発(Research)
Since May 2026·2 on team

QOPT-AKANE — Quantum-Inspired Combinatorial Optimization

An internal R&D solver that evaluates QAOA in classical simulation and combines Tabu Search with a Simulated Bifurcation Machine (SBM) for quantum-inspired combinatorial optimization. Targets Max-Cut, QUBO, and portfolio selection, benchmarked against classical metaheuristics.

≤512 変数 (研究用途)対象問題サイズ
quantumoptimizationqaoa+3
View Demo
Quantum Computing研究開発(Research)
Since May 2026·1 on team

QFAULTTOL-SHIRAITO — Fault-Tolerant Quantum Readiness Study

An internal study using Azure Quantum Resource Estimator, stim, and surface-code simulators to estimate physical-qubit count, runtime, and required error rates for representative algorithms (Shor, QPE, quantum chemistry) under fault-tolerant quantum computation (FTQC). This is a long-term technology-readiness assessment, not real-hardware execution.

Shor / QPE / 量子化学対象アルゴリズム
quantumftqcsurface-code+2
View Demo
AI / Machine Learning研究開発(Research)
Since Apr 2026·1 on team

Kaminari — Neuromorphic Inference Runtime

A neuromorphic inference runtime that executes spiking neural networks (SNNs) on Intel Loihi 2 and BrainChip Akida simulators against spike-encoded sensor data. Aims for milliwatt-class classification of sparse inputs from event-based cameras and ultrasonic sensors. Internal R&D prototype.

98.1%N-MNIST 精度baseline
neuromorphicSNNedge+1
View Demo
AI / Machine Learning研究開発(Research)
Since Apr 2026·1 on team

Asagiri — Federated Learning for Medical Imaging

A federated-learning framework for medical imaging (chest X-ray, dermoscopy) that trains on local GPUs at each site rather than centralizing data. Uses Flower.ai as orchestrator, Opacus for DP-SGD, and secure aggregation for differential privacy. Internal R&D prototype validated only on public datasets — no external medical-institution agreements are in place.

0.847集約 AUCNIH sim
federated-learningprivacymedical-ai+1
View Demo
Automation / Developer Platform開発中(In development)
Since Apr 2026·1 on team

Oboro — Spatial Computing SDK

A spatial-computing SDK that targets Apple Vision Pro, Meta Quest 3, and WebXR browsers from a single scene-graph definition. Thinly abstracts ARKit / Meta XR SDK / WebXR Device API, with Yjs CRDTs syncing spatial-anchor state across multiple devices. Internal R&D prototype.

-38%シーン記述削減vs 単体SDK
spatial-computingXRCRDT+1
View Demo
Automation / Developer Platform開発中(In development)
Since Apr 2026·2 on team

Hibiki — Autonomous Agent Swarm Coordinator

An internal coordination platform built on the A2A (Agent-to-Agent) protocol and MCP (Model Context Protocol), running multiple LLM agents over a shared task queue, memory, and tool-execution surface. NATS JetStream is the message bus, Redis the short-term memory, and SQLite the durable log. R&D prototype.

96.4%タスク完了率20 agents
agentsA2AMCP+1
View Demo
Infrastructure / DevOps開発中(In development)
Since May 2026·2 on team

Kitsune — Edge AI Toolkit

An edge-AI toolkit that unifies the build pipeline and OCI distribution flow for NVIDIA Jetson (Orin Nano / AGX), Raspberry Pi 5, and Amazon Fire TV Stick. TensorRT-LLM, llama.cpp, and whisper.cpp ship from a shared manifest. Internal R&D prototype.

38 tok/sJetson Orin NanoLlama3.2 3B Q4
edge-aijetsonraspberry-pi+2
View Demo

Ready to bring your project to life?

We will propose the ideal IT solution tailored to your business.

Discuss Your Project