AI agents that run parts of your business — not tools you have to learn.
I design and deploy voice receptionists, multi-agent backends, and process automation for small businesses and clinics — built with LangGraph and FastAPI, wired into the tools you already use.
Built with
Five kinds of systems, built to work together
Each one solves a specific bottleneck. Most clients start with one and let it prove itself before adding the next.
AI Voice Receptionists
Custom conversational ElevenLabs voice agents that handle inbound calls, qualify leads, book appointments, and sync data straight to your CRM.
Multi-Agent Orchestration
Complex business workflows managed using LangGraph StateGraph orchestration. Agents coordinate, share context, and run process loops with human-in-the-loop validation.
Backend & API Systems
Production-ready FastAPI microservices integrated with Supabase for real-time database, authentication, and secure row-level data security.
Business Process Automation
End-to-end background processes that run autonomously — reading emails, parsing PDFs, generating reports, and notifying Slack or WhatsApp channels.
Lead Generation Systems
Autonomous scrapers and qualifier agents that target prospects, score them using LLMs, draft highly-personalized reaches, and update dashboards.
Inside the home renovation agent
This is the real orchestration behind my production system for home renovation contractors — a voice agent handles the call, a LangGraph orchestrator routes it, and three sub-agents do the work in parallel, all synced through one shared database.
Deployed on Railway · production system, not a mockup.
github.com/Spectral-Soul/Home_renovation_backendMy core orchestration stack
Strip away the voice layer and this is what every project is actually built on — a LangGraph orchestrator coordinating a LangChain tool-calling agent, Pydantic validating every input and output, and Supabase holding the shared state. Same pattern, different front door.
Same orchestrator, different agents — this is the pattern behind every system on this page.
Talk to my voice receptionist
A live ElevenLabs agent, running the same architecture I ship to clients. Test latency, tone, and how it handles a real request.
How it works
Make sure your mic is on, click the widget below, and speak naturally. The agent qualifies mock leads and books simulated appointments.
Requires microphone access. Runs directly in this page.
Work & projects
Systems I've shipped, in the order I built them — from data integrations to production agent loops.
Customer Support AI Agent
Developed a LangChain SQL database agent deployed on Railway, backed by Supabase, featuring a Vercel-hosted dashboard. Automates 75% of routine order status and customer service queries.
Institute Multi-Agent System
Built a collaborative LangGraph system orchestrating distinct sub-agents: Email Agent (communications), Attendance Agent (processing check-ins), and Fees Agent (automated invoicing).
E-Commerce Operations Agent
Designed a LangGraph architecture implementing parallel-write operations to simultaneously synchronize inventory status and dynamic pricing engines based on demand signals.
EV Parking System
Engineered an IoT-friendly Python system for managing electric vehicle parking logistics, sensor integrations, and charge timing coordination.
AI Receptionist Business — Current Focus
Architecting, tuning, and deploying production ElevenLabs Conversational Voice Agents. Helping local businesses and healthcare clinics scale calls without overhead.
Systems that handle tasks, not just talk.
I specialize in production-grade AI applications — LangGraph state machines and ElevenLabs voice streams that automate repetitive business processes. I operate out of Puducherry, India, building software that replaces manual steps for small businesses and clinics.
Questions worth answering upfront
Direct answers. No fluff, just what you actually need to know.
Let's talk about what to automate first.
Tell me about the repetitive parts of your operation. I'll tell you honestly whether an agent is the right fix — and what it would take to build.