Service
AI Agent & Voice Agent Developer
I build agents that do a specific job inside a real business, not demos. The clearest example is the AI calling agent set I built for a European property management agency: it takes first contact on inbound enquiries, qualifies the lead, books a viewing into the calendar and hands off to a human when it should. That client is under NDA, but the pattern generalises.
- Voice agents for inbound enquiry handling and outbound follow-up
- Scheduling agents wired into real calendars with conflict handling
- Research and enrichment agents that work a list unattended
- Document agents that read, extract and route at volume
- Support agents with escalation paths to a human
- Internal agents over private company data via MCP servers
- LangGraph for stateful, multi-step agents that survive mid-run failures
- LangChain for tool calling and retrieval over your documents
- Claude and OpenAI APIs, chosen per task on cost and latency
- ElevenLabs for voice synthesis on calling agents
- Hermes, OpenClaw and Grokbot for self-hosted agent runtimes tailored to a niche
- Model Context Protocol servers connecting agents to private data
Related
Other things I do
Proof
What clients say
Handled everything from email management to hiring processes. Knowing he had the operations covered allowed me to focus purely on growth.
Managed our luxury property listings and handled customer troubleshooting, automation, and project management flawlessly. Incredible attention to detail.
Managed my email, project pipelines, invoicing, and even delivered a custom Squarespace website design. He takes execution to the next level.
Supported our growth through targeted lead generation and meticulous online research. His data enrichment built us a robust sales pipeline.
Questions
What people ask about this
Do AI voice agents actually work on real calls?
For narrow, well-defined jobs, yes. Qualifying an inbound enquiry and booking a viewing works reliably. Open-ended conversation where anything might come up does not, and I build in a handoff to a human rather than pretending otherwise. The agent knowing when to stop is most of the engineering.
What happens when the agent gets something wrong?
It escalates. Every agent I build has a confidence boundary and a human handoff path, plus logging so you can review what it did. An agent that fails loudly is worth far more than one that fails quietly and confidently.
Can the agent use our internal data?
Yes, through MCP servers or retrieval over a controlled document set. For clients with data sensitivity I run this self-hosted so nothing is sent to a third-party service you have not approved.
What does an agent build cost?
Agent builds start around USD 1,000 and scale with how many systems the agent has to touch. Voice agents cost more than text because the call infrastructure and testing take longer. I scope after understanding the specific job, not from a price list.
Get in touch
Tell me what you need.
Describe the problem rather than the solution. I will tell you honestly what it takes to fix.
WhatsApp · fastest
+91 98756 02929
Email
sentu1993@gmail.com
LinkedIn
souptik-guha-roy
Typical reply time
Within one business day