Keywords: RAG, Agentic RAG, LangGraph, Codex SDK, Context Engineering

• Development of AI agents that leverage accumulated knowledge in diverse formats
• Improving question-answering accuracy based on in-house technical reports and past interaction logs
• Development of systems that use AI to automatically and quantitatively evaluate AI performance
• Building systems for continuous automated evaluation of AI agent performance from multiple perspectives
• Interface development and deployment to cloud environments
• Development of features such as external system integration and authentication

Keywords: AI Agent, MCP, OpenAI Agents SDK, A2A, AGENTS.md, Skills

• Development of AI agents to streamline R&D operations
• Building R&D-specialized AI capable of automated literature surveys and public database exploration
• Operational support for development workflows based on parallel use of coding AI
• Providing expertise on running coding AI in parallel within your development environment
• Integration of AI agents with existing applications
• Development of APIs that enable AI agents to interact with your applications

Keywords: Local LLM, SLM, Evals, Safeguard, vLLM, SGLang, Anonymization

• Implementation and tuning of local LLMs running in on-premises environments
• Optimizing inference speed of local LLMs serving a large number of concurrent users
• Development of a real-time minute-taking system running in on-premises environments
• Generating minutes from meeting audio using locally running speech recognition models
• Generating externally shareable training data through anonymization of unstructured data
• Anonymizing personal information by masking names, locations, and addresses in unstructured data

Keywords: GRPO, DPO, SFT, Mamba, SSM, MoE, DSPy, GEPA

• Replication and extension of state-of-the-art research papers
• Development of large-scale ETL pipelines for training data collection
• Implementation of text-input regression models using large language models
• Automated generation and tuning of prompts for AI

Automated Improvement of Mathematical Algorithms through Evolutionary Computation and LLMs

We built a system that autonomously improves mathematical algorithms by combining large language models with evolutionary computation. A CLI-based AI coding agent iteratively modifies and evaluates the algorithms.

Through 48 hours of automated execution, the system achieved performance exceeding manual tuning (Accuracy: 0.914 → 0.956).

Development Environment & Technical Fields: Automated Algorithm Design, CLI-based AI Coding Agent, Evolutionary Computation

Automated Planning System Using Iterative LLM Refinement

We built an automated planning system for software development by iteratively refining the output of a large language model.

Given a software requirement specification as input, this system interactively confirms the details of the requirements on the GUI and decomposes them into the tasks necessary to achieve the goal. The system is actively used in our day-to-day operations.

Development Environment & Technical Fields: Large Language Models, AI Agents