# TurinTech AI > Artemis improves the performance of models, agents and code you already run by testing multiple optimization strategies and delivering measurable gains. TurinTech builds Artemis, an AI optimization platform. Give it a project and the metric it is judged on, or let it scan the codebase against your own rules to find one; it searches for better versions of the code, models or agent settings, benchmarks each one on your own hardware, and hands back the winners as pull requests. The pages below are the canonical sources; each page description is the page's own. Product documentation lives on a separate host with its own index. The docs describe Artemis by its four modules, Build, Maintain, Discover and Optimise; on this site, "Discovery" is the Discover module and "Scan" is Maintain. ## Product - [Artemis: optimize code, models and AI agents](https://www.turintech.ai/product/overview): Artemis by TurinTech optimizes the models, agents and code you already run. Discovery tests multiple approaches against real metrics and keeps the best. - [Artemis in your workflow with your coding agent](https://www.turintech.ai/product/sdlc): Whichever coding agent writes the code, Artemis searches for its best version on models you choose, proves the win on your hardware, and repeats it on every PR. - [Artemis deployment: VPC, on-premise, air-gapped](https://www.turintech.ai/product/deployment): You decide where Artemis runs: managed by us, your cloud account, your data center, an air-gapped network or one workstation, using only models you approve. - [Artemis ROI calculator](https://www.turintech.ai/artemis-roi-calculator): Estimate the engineering savings and business outcome uplift Artemis could deliver for your production AI workloads. - [Artemis documentation](https://docs.artemis.turintech.ai): User documentation for deploying and using Artemis, including Discovery tutorials for driving Artemis through the CLI with a coding agent. - [Artemis documentation index (llms.txt)](https://docs.artemis.turintech.ai/llms.txt): The documentation host's own llms.txt: architecture, deployment, features and release notes. - [Artemis quick overview](https://docs.artemis.turintech.ai/getting-started/quick-overview): The four modules of Artemis: Build, Maintain, Discover and Optimise. - [Artemis CLI](https://docs.artemis.turintech.ai/features/artemis-cli): Command-line interface for the Artemis AI code platform. - [Artemis Skills](https://docs.artemis.turintech.ai/features/artemis-agent-skills): Claude Code plugins that guide you through setting up and configuring Artemis from your terminal. - [Artemis Runner](https://docs.artemis.turintech.ai/features/artemis-runner): Validates your code by running builds, tests and benchmarks on your own machine. - [LLMs in Artemis](https://docs.artemis.turintech.ai/architecture/llm-usage): Built-in LLM providers, plus bring your own LLM through a custom provider profile. - [Deployment options](https://docs.artemis.turintech.ai/deployment/deployment-options): The five supported ways to access Artemis. ## Use cases - [LLM inference optimization on existing hardware](https://www.turintech.ai/use-cases/ai-inference): You don't need more GPUs or CPUs. Artemis optimizes the inference stack you already run (vLLM, TensorRT-LLM, SGLang) so the same hardware serves more requests. - [AI agent optimization for cost and task success](https://www.turintech.ai/use-cases/ai-agents): Tune the prompts, tool descriptions and settings around your agent, then keep the versions that cost less and complete more tasks successfully. - [Latency-critical code optimization](https://www.turintech.ai/use-cases/latency-critical-systems): Trading engines, bidding paths, control loops, live APIs: an answer only counts if it lands in budget. Artemis made a bank's quant library 29% faster. - [SQL and data pipeline cost optimization](https://www.turintech.ai/use-cases/data-and-sql): Your dashboards and pipelines stay as they are. Artemis rewrites the SQL underneath, checks every version returns the same results, and keeps what bills less. - [Routing, scheduling and allocation optimization](https://www.turintech.ai/use-cases/decision-algorithms): Improve routing, scheduling and allocation without replacing the systems your teams use. Artemis finds better solutions within your objectives and constraints. - [Improve ML model accuracy on your own metric](https://www.turintech.ai/use-cases/ml-models): Your data stays as it is. Test hundreds of improvements to your existing model and keep only the ones that perform better against the metric you already use. ## Measured results - [2.07× LLM inference throughput (vLLM, Qwen 3.6 35B, Intel Xeon 6)](https://www.turintech.ai/use-cases/ai-inference): 38 → 79 output tokens/s with outputs unchanged, measured on vLLM on Intel Xeon 6, in a build Intel formally approved; published in Intel's partner brief. - [+40% LLM inference throughput (TensorRT-LLM, Nemotron-3-Nano-30B-A3B, NVIDIA A100)](https://www.turintech.ai/use-cases/ai-inference): 1,203 → 1,690 output tokens/s, 30% lower median time per output token and 46% less weight memory, measured on Nemotron-3-Nano-30B-A3B on one NVIDIA A100 80 GB, TensorRT-LLM PyTorch backend at concurrency 32, BF16 baseline versus W8A16; merged into NVIDIA's repository as TensorRT-LLM pull request #15550. - [4.9× faster cross-attention in OpenAI Whisper](https://www.turintech.ai/open-source): 424 → 87 µs on the attention call and 19% faster end-to-end transcription, measured on Whisper large-v3, fp16, beam size 5, on an NVIDIA RTX 3090 with FlashAttention; word error rate unchanged; merged upstream as openai/whisper pull request #2812. - [29% faster quant library](https://www.turintech.ai/use-cases/latency-critical-systems): 29% faster runtime on a leading bank's proprietary quant library, on CPU. - [14% less time on a hashing hot path](https://www.turintech.ai/use-cases/latency-critical-systems): Measured on the hashing path of a Solana validator client at a leading global trading firm, on AMD EPYC 9B45. ## Evidence - [Open-source contributions from Artemis](https://www.turintech.ai/open-source): We build on llama.cpp, Whisper, TensorRT-LLM, OpenVINO, QuantLib and others. When Artemis finds a fix, an engineer opens a pull request with the measurement. - [Optimization research behind Artemis](https://www.turintech.ai/research): Optimization is a research problem. Eight years of published work on search, evaluation and validation for optimizing production systems, built into Artemis. ## How to work with us - [Book an Artemis demo](https://www.turintech.ai/book-a-demo): One metric, one repo, one benchmark is all a pilot needs. Tell us which workload you want faster, cheaper or more accurate, and we reply in one business day. - [Get started with Artemis](https://www.turintech.ai/get-started): Our engineers optimize a system you run, in your environment, and return benchmarked pull requests. Free for selected research and open-source projects. - [AI for Good](https://www.turintech.ai/ai-for-good): Optimized code saves energy, keeps hardware alive longer and lets research run sooner. Research and open-source projects can submit a repo; fixes go upstream. ## Company - [TurinTech AI](https://www.turintech.ai/): Artemis improves the performance of models, agents and code you already run by testing multiple optimization strategies and delivering measurable gains. - [About: the UCL spinout behind Artemis](https://www.turintech.ai/about): TurinTech is a UCL spinout building AI that makes production systems measurably better. Every improvement is benchmarked before it counts. - [Blog: engineering write-ups and news](https://www.turintech.ai/blog): Everything TurinTech publishes, newest first — engineering write-ups, product news, partnerships and press coverage, in one place. - [Newsroom](https://www.turintech.ai/news): Launches, partnerships and press coverage — everything TurinTech has announced, in one place. - [RSS feed](https://www.turintech.ai/feed.xml): Blog and news, newest first. - [Atom feed](https://www.turintech.ai/atom.xml): The same articles as Atom. - [Sitemap](https://www.turintech.ai/sitemap.xml): Every public page and article on this site.