I build AI agent systems that replace manual business operations. I run South Arc Digital, a technical automation agency serving clients in the US and India.
Claude Code CLI is my primary development environment.
- Custom MCP Server (35+ tools) — Purpose-built Model Context Protocol server for a production textile ERP. 79-column MS SQL schema with embedded gotcha documentation for schema inconsistencies. One of the earliest production MCP implementations for a real business.
- Multi-Agent Orchestration (5 LLMs) — Orchestrates Claude Opus, GPT-5.4, Grok 4.1, DeepSeek V3.2, and Gemini via OpenRouter. 6-phase pipeline: Planning → Distribution → Parallel Execution → Peer Review → Synthesis → Voting. Swappable skill configs via YAML.
- ReAct Agent + WhatsApp Bot — Multi-step reasoning agent generating dynamic SQL against a legacy ERP. Up to 10 reasoning steps per query with self-correction. Deployed as a WhatsApp bot — 7,100+ SKUs searchable via natural language.
- Dual-Tier LLM Intelligence Platform — Claude Haiku (fast screening) → Gemini Pro (deep analysis). Monitors Reddit, Twitter, job boards. Per-model cost tracking with auto-pause on budget thresholds. Generated a 10,044-datapoint VoC intelligence library.
- 95 GB Data Migration — 43,926 products, 15,463 matched image URLs, 18 batch imports for a US luxury furniture retailer. Python ETL with checkpoint resumability. Zero data loss.
- Cylinder Rotation Analytics — Connected TrackAbout + Zoho Books for an industrial gas company. 6 dashboard reports, WhatsApp alerts, 15-min sync. Ongoing maintenance contract.
- Textile Business Intelligence — AI WhatsApp bot + 17-report dashboard over legacy MS SQL ERP. ReAct agent, custom MCP server, 4-tier auto-sync, GraphQL API, 279 tests.
- ERPNext Implementation — ERP deployment + AI integration for smart document ingestion for a polymer manufacturer.
| Capability | Context |
|---|---|
| Custom MCP Server (35+ tools) | Production — Textile ERP |
| ReAct Agent (multi-step reasoning) | Production — WhatsApp bot |
| Multi-agent orchestration (5 LLMs) | Internal tool |
| RAG pipeline (embeddings → vector search → generation) | POC — Legal documents |
| NLP order parsing (multilingual Hindi-English) | POC — FMCG distribution |
| Multi-factor business scoring algorithms | Production — Textile analytics |
| Dual-tier LLM filtering (fast + deep) | Internal tool |
| Hybrid fuzzy search (pg_trgm + LLM fallback) | Production — Textile search |
| Voice transcription + entity extraction | POC — Field sales tool |
| Natural language → SQL with self-correction | POC — Manufacturing analytics |
AI & Agents: Claude Code CLI, LangGraph, MCP, OpenAI API, Anthropic Claude API, Google Gemini, DeepSeek, Grok, OpenRouter, Qdrant (vector DB)
Backend: Python (FastAPI), Node.js (Express), TypeScript
Frontend: Next.js, React, TailwindCSS, shadcn/ui
Databases: PostgreSQL, MongoDB, MS SQL, SQLite, Qdrant
Integrations: REST APIs, Webhooks, OAuth, Twilio (WhatsApp/Voice), Google Workspace APIs, Zoho APIs, Shopify Admin API, n8n
Infra: Railway, Vercel, GCP (Cloud Run, GCS, Firebase), Docker, Cloudflare Workers
Most of my work is for paying clients under NDA. I can't open-source production business systems containing real customer data, API keys, and proprietary business logic.
What I can share:
- 📄 Portfolio and case studies at southarcdigital.com
- 📊 1,100+ contributions in the last year — I ship daily
- 🎥 Walkthroughs and demos available on request
Law degree from National Law School of India University (India's #1 ranked law school) → International arbitration at India's largest law firm → In-house counsel at ITC Limited (Fortune 500 FMCG) → Taught myself to code → Founded South Arc Digital.
I understand business operations, compliance, and regulated industries from the inside. That domain knowledge shapes how I architect systems — I don't just build the tech, I understand the problems it solves.