Votee offers a low‑code platform for building and deploying AI agents that are fine‑tuned to local languages and cultural nuances, achieving up to 90 % accuracy in localized LLM evaluation. Its visual agent builder, high‑throughput SageMaker training, and proprietary benchmarking‑as‑a‑service enable enterprises—especially in regulated fintech and property‑tech sectors—to quickly create compliant, culturally aware multilingual agents while reducing training costs and time.
Funding
Funding not disclosed
Founders
Product
Problem
Enterprises deploying large language models often face poor performance in low‑resource languages and cultural contexts, leading to inaccurate responses, mis‑interpretations, and costly AI failures. Additionally, there is a lack of rigorous, domain‑specific benchmarking tools to evaluate model accuracy and compliance before production deployment.
Solution
Votee provides a platform for building and deploying AI agents that are fine‑tuned to local languages and cultural nuances, delivering up to 90 % accuracy in localized LLM evaluation. The solution includes a visual agent builder powered by Google ADK, enabling rapid creation of custom agents without deep engineering effort. Votee’s proprietary benchmarking‑as‑a‑service evaluates both proprietary and open‑source models on professional knowledge, cultural relevance, and linguistic detail such as Cantonese phonology. The platform leverages high‑performance SageMaker HyperPod clusters and a MongoDB‑based data pipeline to train, fine‑tune, and extract insights from multilingual document data at reduced cost and increased speed. Compliance‑focused agents are also offered to meet regulatory requirements in financial and property‑tech sectors.
Target Audience
Primary customers are enterprises in fintech, property management, and other regulated sectors that require accurate, culturally aware AI agents for local markets such as Hong Kong, Singapore, Malaysia, and Vietnam.
Features
- Visual AI Agent Builder (Google ADK) for low‑code creation and orchestration of multilingual agents
- High‑throughput LLM training on SageMaker HyperPod clusters delivering up to 10× cost and performance improvements
- Cantonese‑specific benchmark (HKCanto‑Eval) measuring professional knowledge, cultural nuance, and phonological accuracy
- Multilingual data extraction pipeline (“Vision Model”) that transforms document data into structured AI‑ready formats using MongoDB
- Benchmarking‑as‑a‑Service offering evidence‑based model selection and performance validation for enterprise use cases
- Compliance AI agents designed for fintech and property‑tech regulatory environments (e.g., HKMA FSS 3.1 pilot)
- Agentic training approach that enhances LLM alignment and data operation efficiency (95 % agentic approach)