Hire · Freelance & consulting
Hire an AI agent developer who builds agents that hold up in production
I build LLM agents that face real users: conversational agents on LLaMA and OpenAI APIs for clients at Shivohini TechAI, AI voice agents at Zuneko Labs, and Agentic Honey-Pot, a multi-persona agent that engages scammers in real time. The model is the easy part; context, state and failure handling are the job.
Problems I solve with this stack
The agent forgets earlier turns or blows the token budget.
Explicit context management: decide what to carry forward, summarise or drop on every turn. That was the core of the conversational agents I built on LLaMA and OpenAI APIs at Shivohini TechAI.
One prompt can't handle every kind of conversation.
Strategy switching: distinct personas or modes the agent moves between as the conversation changes, like the four strategies (passive, vulnerable, engaged, stalling) in the Agentic Honey-Pot persona engine.
Replies are slow and block the rest of the system.
An async backend: FastAPI receives the event, Celery and Redis run inference off the request thread, and the reply goes out when it's ready.
Nobody can explain afterwards why the agent said what it said.
Log each turn: the input, the context that was sent, the strategy or tool chosen and the reply. A bad answer can then be replayed and fixed from evidence instead of by guessing at the prompt.
What you get
- An LLM agent on OpenAI, LLaMA or Groq-hosted models, with explicit context and state management.
- A FastAPI backend with Celery + Redis so inference never blocks requests.
- Channel integration: the WhatsApp Business API, voice, or your own app.
- Guardrails: structured flows for critical steps and a human hand-off path.
- Logs of what the agent did, for review and tuning.
How a AI agents / LLM engagement runs
1. Define the agent's job
Agree what it may do on its own, which steps need a structured flow, and when it hands the conversation to a human.
2. Build a thin version on one channel
A FastAPI backend, one model, explicit context management and one channel, such as WhatsApp or voice, before anything more is added.
3. Test on realistic conversations
Replay transcripts, including awkward ones such as off-topic messages and repeated questions, read where the agent goes wrong, then adjust prompts, flows and guardrails.
4. Make it async and hand over
Move inference to Celery + Redis workers, add logging, and hand over the prompts, flow definitions and a runbook.
Frequently asked questions
What's the difference between an AI agent and a chatbot?
A chatbot answers; an agent also keeps state and takes actions, such as switching strategy, calling tools or handing off to a human. Most real projects need a mix of both.
Which LLM do you build on?
Whatever fits the constraints. I've built on OpenAI and LLaMA APIs, and Agentic Honey-Pot runs on models served through the Groq API; cost, latency and data rules decide it.
Can the agent hand a conversation over to a human?
Yes, and wherever a wrong answer costs money it should. I design the hand-off path and log what the agent did, so a person can pick up the conversation with context.
How do you test an agent before real users see it?
I replay realistic conversations and read the transcripts. Automated checks can cover the structured steps, but open-ended replies still need a person reading them.
Can the agent call our internal tools or APIs?
Yes. Tools are exposed as validated function calls with limits on what each can do. Critical actions go through a structured step or a human approval instead of free model judgement.
Do you work with international clients?
Yes. I'm based in Pune, India, and work with teams in India and remotely worldwide. Communication is async-first with scheduled calls when needed.
What does an engagement look like?
A short call to scope the problem, a written scope (what will be built, what won't, and how we'll know it works), then small milestones you can review, with the code in your repository from day one.
More questions? Ask directly
Last updated
Need help with AI agents / LLM?
Book a free 15-minute call below — describe the problem, I'll tell you honestly whether I'm the right person for it.
How I work with clients