Research at Salesteq
We're not building wrappers around existing models. We're doing original research to create autonomous systems that can run commercial operations end-to-end: reliably, continuously, and at scale.
Agentic Systems
Autonomous agents that plan, act, and learn
Traditional automation follows scripts. Our agents reason about goals, decompose them into steps, execute across multiple tools and data sources, and adapt when things change. We're developing architectures for long-running agent workflows that maintain coherent state across days and weeks of continuous operation.
Our work spans multi-agent coordination, tool-use optimization, error recovery strategies, and context management for enterprise-grade reliability. Every agent is designed to operate in production business environments where mistakes have real cost.
AI Model Intelligence
Making AI models work in high-stakes business contexts
Frontier AI models are powerful but unreliable out of the box. We research how to ground them in structured enterprise data (CRMs, pipelines, financial records) so they produce accurate, actionable output rather than plausible-sounding guesses.
This includes work on retrieval-augmented generation over heterogeneous business data, optimization strategies for domain-specific tasks, confidence calibration, and multi-model orchestration where different AI models handle different parts of a complex workflow based on their strengths.
Kol: Speech and Voice Intelligence
Natural conversation between agents and people
Led by a co-founder whose speech recognition engine has been downloaded millions of times worldwide, Kol is our proprietary voice platform. Real-time, low-latency speech systems that enable AI agents to have natural phone and video conversations.
We work on end-to-end speech pipelines including voice activity detection, streaming transcription, speaker diarization, sentiment analysis, and neural text-to-speech, all optimized for the nuances of business communication where tone, timing, and precision matter.
The Future of Work
Engineering a better workplace for everyone
The goal isn't to eliminate people. It's to free them. We research how AI agents and human teams work together most effectively, so people spend their time on judgment, creativity, and relationships instead of data entry, follow-ups, and administrative work.
This means studying human-AI collaboration patterns, designing transparent agent behavior that builds trust, creating feedback loops where human expertise improves agent performance, and measuring what actually makes work meaningful when routine tasks are automated away.
Our research team
Salesteq's research is led by Dr. Milan Jelisavcic (PhD in AI, inventor of universal transfer learning algorithms) and Dr. Nickolay Shmyrev (whose speech recognition engine has millions of downloads worldwide). We're based in Zug, Switzerland.
Meet the team →