| --- |
| title: README |
| emoji: 🧭 |
| colorFrom: gray |
| colorTo: indigo |
| sdk: static |
| pinned: false |
| --- |
| |
| # Solvren AI |
|
|
| AI forward deployment for mid-market regulated teams. One production AI workflow per engagement, built inside your own cloud in 6 to 8 weeks. |
|
|
| ## What we do |
|
|
| - Our engineers embed with your team and build one production AI workflow per engagement. |
| - The build runs inside your own AWS, Azure or GCP account, on models you already approved. |
| - You keep the code, the infrastructure and the runbooks. |
| - Systems we have built: an AI skills registry with security scanning, agentic payments on Amazon Bedrock AgentCore, and Class 1 Decider, a small routing layer in front of LLM calls. |
|
|
| We work with biotech, healthcare, defense, legal and fintech teams. Founded in 2026, based in San Diego. |
|
|
| ## Open models & datasets |
|
|
| | Artifact | What it is | Result | |
| |---|---|---| |
| | [class1-router-qwen3-0.6b-mlx](https://huggingface.co/SolvrenAI/class1-router-qwen3-0.6b-mlx) | Qwen3-0.6B LoRA request router, MLX | 0.88 accuracy on 50 hand-written requests (untuned Qwen3-4B: 0.90), a third to half the p50 latency | |
| | [class1-router-synthetic](https://huggingface.co/datasets/SolvrenAI/class1-router-synthetic) | Router training data, no real PII or PHI | 1,536 template rows + 50 hand-written test rows | |
| | [bge-base-en-v1.5-hipaa](https://huggingface.co/SolvrenAI/bge-base-en-v1.5-hipaa) | HIPAA retrieval embeddings | NDCG@10 0.907 vs 0.834 for the base model, 99 held-out queries | |
| | [hipaa-retrieval-pairs](https://huggingface.co/datasets/SolvrenAI/hipaa-retrieval-pairs) | Query-passage pairs over 45 CFR Parts 160 and 164 | 912 pairs over 304 regulation chunks | |
| | [hipaa-fda-rag-eval](https://huggingface.co/datasets/SolvrenAI/hipaa-fda-rag-eval) | Synthetic RAG eval QA over HIPAA 45 CFR 164 and FDA device-cybersecurity guidance, plus a 345-chunk corpus | 162 QA rows; reference answers are noisy (see below) | |
|
|
| Read the numbers with care: |
|
|
| - These are demos trained on synthetic or LLM-generated data. |
| - The test sets are small (n=50 and n=99), so one error moves the router's accuracy by 2 points. |
| - Router latency was measured on one shared laptop that was running other jobs, so it is approximate. |
| - In the RAG eval set, about 28% of spot-checked reference answers are partial or wrong, and two-thirds of questions are yes/no with answers skewed heavily to "Yes". Use it mainly for retrieval recall; the card has details. |
| - On the HIPAA eval the fine-tuned embedding model also scores ahead of bge-large-en-v1.5 (0.864) and snowflake-arctic-embed-m-v1.5 (0.865), with no meaningful change on BEIR SciFact. |
|
|
| Each model card covers the eval setup and its limitations. |
|
|
| ## Code |
|
|
| - [solvren-ai/class1-router](https://github.com/solvren-ai/class1-router): data generation, LoRA training and evaluation for the router |
| - [solvren-ai/domain-embedding-finetune](https://github.com/solvren-ai/domain-embedding-finetune): local embedding fine-tuning pipeline |
| - [solvren-ai/rag-eval-generator](https://github.com/solvren-ai/rag-eval-generator): synthetic RAG evaluation set generation |
|
|
| ## Compliance and partners |
|
|
| We build for teams working under HIPAA, SOC 2, ITAR and CMMC. Compliance and any certification or authorization stay with your organization and your auditors. |
|
|
| Solvren AI is an OpenAI Select Partner and an AWS Partner Network member. |
|
|
| ## Citing our work |
|
|
| Each model and dataset card has a Citation section with a BibTeX entry and upstream credits. Please cite the specific artifact you use. |
|
|
| ## Contact |
|
|
| [solvren.ai](https://solvren.ai) · [Contact](https://solvren.ai/contact/) · [nitya@solvren.ai](mailto:nitya@solvren.ai) |
|
|
