Phonx AI
Production voice AI for US insurance enrollment: sub-second turns, compliance-aware reasoning and CRM write-back.
Unwind AI
Unwind AI / AI consulting
AI consultancy and AI development
Unwind AI is an AI consultancy run by engineers. We find the workflow where AI pays off, design the system, build it into the tools your team already uses and hand it over. Voice agents, knowledge systems that cite their sources, and automation that sends unsure cases to a person.
AI consulting services
A written plan for where AI helps your business, what to build first and what it would take, or an honest note if it is not worth building.
Workshops, architectureAgents that read, decide and act inside your systems, with clear boundaries and a person in the loop where it matters.
LLMs, tools, guardrailsPhone agents that answer and call under a one-second turn budget and write every outcome into your CRM.
Twilio, Groq, ElevenLabsRetrieval over your documents and data, with knowledge graphs, compliance rules at retrieval and answers that cite their sources.
Neo4j, embeddings, FastAPIEmail, document and support workflows automated end to end, with confidence-based escalation to your team.
OpenAI, Zapier, PythonHands-on review and engineering for an AI system you already run: latency, cost, accuracy and observability.
Reviews, pairing, fixesHow we work
Constraints, stakeholders, risks and hidden assumptions, mapped before a line of code is written.
Interfaces, data flows and AI boundaries defined first, so every later decision is faster and less risky.
Modular from day one, with the logging and monitoring that let a system be improved in production.
Validation continues past launch, because the best systems improve after deployment.
Production voice AI for US insurance enrollment: sub-second turns, compliance-aware reasoning and CRM write-back.
A Neo4j vector graph that enforces compliance rules at retrieval, in production in insurance.
Live mandi prices and a 14-day forecast that reports its own uncertainty, live at kycagri.com.
Read the full write-ups in our case studies.
Questions
It starts with a free 30-minute call about the workflow and its constraints. You then get a written scope: what is worth building first, the architecture, and what it would take. If you go ahead, we build a pilot on one high-value workflow, measure it against how it runs today, then roll it out.
Both. The engineers who scope your system build it, and hand over code, configuration and documentation your own team can run.
Voice AI agents for phone calls, knowledge agents and RAG systems that cite their sources, workflow automation for email and documents, LLM integrations into existing products, and data and forecasting pipelines.
Whatever fits the job: OpenAI and open-weight models such as Llama, Groq for low-latency inference, ElevenLabs for voice, Neo4j for knowledge graphs, and Python and FastAPI services, built around the tools you already run.
Access and compliance rules are enforced inside the system, with audit logging where it matters. Sensitive deployments can run inside your own environment; a previous insurance deployment runs on-premises with encryption at rest.
Then we will say so. Some problems are better solved with a rule, a form or a process change, and the scoping step is designed to find that out early.