Services / 01
Custom AI agents, multi-agent orchestration and agentic automation for startups and SMBs. Claude, GPT-4, LangChain, MCP. Voice and video interfaces. Real production systems — PatFace, Iklavya — not demos.
Single-purpose and multi-purpose agents with planners, critics, tool use, memory and schema-validated outputs. Built to replace repetitive decision work, not to impress in a demo.
Coordinator + specialist pattern — a planner routes tasks to focused workers, a critic validates. Shipped at scale in PatFace (patent drafting) and Iklavya (AI interviewer).
Realtime agents with ElevenLabs TTS, HeyGen/Simli avatars, streaming ASR. Lip-synced, interruptible, fast enough for live interviews and customer support.
Model Context Protocol servers so Claude and other agents call your tools safely and deterministically. Typed schemas, auditable tool calls, no hallucinated parameters.
Token budgets, retry bounds, output schemas, critic agents, and replay logs. The system fails small and fails visibly.
AWS / GCP / Vercel / Railway deployment. Monitoring with CloudWatch or similar. Ongoing retainer available for iteration and oncall.
LLMs
Anthropic Claude (Opus, Sonnet, Haiku) · OpenAI GPT-4 / GPT-4o · Google Gemini · Ollama · Mistral
Agent frameworks
LangChain · LangGraph · CrewAI · Haystack · custom orchestrators (TS/Python)
Retrieval & memory
Pinecone · ChromaDB · Weaviate · Qdrant · pgvector · Redis
Voice & video
ElevenLabs · HeyGen · Simli · Deepgram · Remotion
Tool protocol
MCP (Model Context Protocol) · OpenAI tool calling · Anthropic tool use · custom REST adapters
Deploy
AWS (EC2, ECS, Lambda) · GCP · Vercel · Railway · Docker · GitHub Actions
patmaster.online
iklavya.in
shipsarthi.com
Prototype
$2,500 – $6,000
Single-purpose agent with one or two tool integrations, deployed for internal use. 1–2 weeks.
Production
$8,000 – $20,000
Hardened agent with guardrails, monitoring, CI/CD, handover docs. 3–5 weeks.
Platform
$25,000+
Multi-agent system with RAG, voice/video, admin, billing, multi-tenant. 6–12 weeks. PatFace / Iklavya scale.
Fixed-price or monthly retainer. NDA and IP-assignment standard. Hourly available on request.
AI agent development is the practice of building autonomous software agents that use large language models (LLMs like Claude and GPT-4) to plan, reason and take actions — calling tools, reading data, writing to systems, and coordinating with other agents. A production AI agent is not a chatbot; it is a bounded system with a goal, a memory, a tool-belt and a critic.
Startups building AI-powered products, SMBs replacing manual workflows, and enterprises integrating LLMs into internal operations. If your business has a repetitive task that involves reading documents, making decisions, and taking actions — an AI agent can probably do it. Typical engagements range from $3,000 MVPs to $30,000 multi-agent platforms.
A single-purpose agent MVP with one tool integration ships in 1–2 weeks. A multi-agent system with RAG, voice, and production guardrails typically takes 4–8 weeks. PatFace (agentic patent drafting) and Iklavya (agentic interviewer) are examples of the larger end.
LangChain, LangGraph, CrewAI, and custom orchestration in TypeScript/Python. For tool use I prefer the Model Context Protocol (MCP) where applicable because it keeps the agent’s interface explicit. LLM providers: Anthropic Claude (Opus, Sonnet), OpenAI GPT-4, Google Gemini.
Every agent has a schema-enforced output (JSON), a token budget per task, a retry bound, and a critic agent that validates before acting. Costs are capped per session. Tool calls are logged and replayable.
Yes. I’ve shipped voice and video agents using ElevenLabs (TTS), HeyGen and Simli (avatars), and streaming ASR for real-time conversation. See the Iklavya case study.
Yes. Typical integrations include CRMs (HubSpot, Salesforce), databases (Postgres, MongoDB, MySQL), SaaS APIs (Slack, Notion, Linear, Zendesk), payment gateways, and custom internal tools via REST or MCP.
Week 1: discovery call, scope, success metrics. Week 2: prototype with real data. Week 3–4: production hardening with guardrails, tests, cost monitoring. Week 5+: deployment and handover. Monthly retainer optional for ongoing development.
Same-week start. Email reply within 24 hours. Written enquiries welcome.