Browse categories
Explore
Fiverr Pro
English
$
USD
I will build custom ai agents using langgraph and crewai
Abdullah Khan
About this gig
Need an AI agent that works in production, not a demo that falls over on the third request?
I build custom AI agents with LangGraph and CrewAI that do real work. They pull data, make decisions, call your tools, and stop to ask a human when they're not sure.
What you get:
A working agent, deployed and wired into your stack: Slack, CRM, Google Sheets, any REST API
Memory that persists across sessions, so it doesn't forget what happened yesterday
RAG over your own documents, so answers come from your data instead of guesswork
Approval checkpoints, so it never takes an action you didn't sanction
Model routing and semantic caching to keep costs down. This cut API latency 60% on one build
Source code and docs. You own all of it.
Most of my work has been mortgage and healthcare, where a wrong output costs someone real money. I build to that standard even when a project doesn't demand it.
Tell me what you want automated. If an AI agent is the wrong tool for the job, I'll say so.
Get to know Abdullah Khan
Abdullah Khan
AI Architect: 5 years, Enterprise RAG Systems, Agents and AWS MLOps
- FromPakistan
- Member sinceJul 2024
- Avg. response time15 hours
- Last delivery1 month
Languages
Urdu, English
I architect Enterprise RAG & Agent systems that cut operational overhead by 40%. From Proof-of-Concept to Production, I build scalable compliant infrastructure.
My focus is on eliminating enterprise bottlenecks: high hallucination rates, latency, and spiralling costs.
I possess deep domain expertise in highly regulated sectors. I design "Human-in-the-Loop" architectures that ensure high-uncertainty decisions for review, ensuring safety and compliance.
If you require a scalable, secure, and high-performance AI architecture, I invite you to contact me to discuss your requirements.
My Portfolio
FAQ
How do you prevent agents from entering infinite loops or hallucinating?
I engineer deterministic state machines using LangGraph, enforcing strict transition logic and "Human-in-the-Loop" checkpoints. This ensures agents execute within defined boundaries, preventing infinite loops and unauthorized actions common in basic autonomous business logics.
What is the strategic advantage of a Multi-Agent System over a single LLM?
A single LLM hallucinates when overloaded with context. Multi-Agent Systems assign specific roles to distinct agents. This "separation of concerns" drastically improves accuracy, reduces latency, and allows for parallel business tasks automation.
Do your agents possess long-term memory and context retention?
Yes. I engineer persistent state management using Redis (Short-term) and Vector Databases (Long-term/RAG). This allows agents to recall past interactions, user preferences, and institutional knowledge, enabling complex, multi-session workflows.
How do you manage token consumption and operational costs?
I implement cost-aware routing architecture. Simple tasks are routed to efficient models (Llama 3/GPT-4o-mini), while complex reasoning uses GPT-4o. I also deploy semantic caching to prevent redundant API calls, reducing operational overhead by up to 40%.
Can these agents perform actions on my internal software?
Absolutely. I build Custom Toolkits (APIs) that allow agents to securely interact with your internal CRM, ERP, or Database. I implement strict OAuth2 authentication and permission layers so agents can only perform authorized actions (e.g., Read-Only vs. Write).
What exactly is included in the Basic package?
One AI agent built for a single defined task, connected to one tool or API of your choice, with conditional logic and an AI model integrated. You get the source code and a short handover doc. If you need several agents working together, that is the Standard package.
What do you need from me to get started?
A clear description of the task you want automated, and access to whatever the agent needs to touch: API keys, a Slack workspace, a CRM, a folder of documents. If you are not sure what access is needed, message me first and I will tell you before you order.
Do I own the code and can I host it myself?
Yes. You get the full source code and it is yours. Everything runs on your own infrastructure and your own API keys, so there is no dependency on me after handover and no monthly fee to keep it running.
How do revisions work?
Basic includes one revision, Standard two, Premium three. A revision covers adjusting behaviour, prompts, logic or integrations within the scope we agreed. Adding a new capability that was not in the original brief is a separate order, and I will tell you which it is.
What if the agent does not do what I expected?
We agree the exact behaviour in writing before I start, so surprises are rare. If the delivered agent does not match that brief, I fix it under your included revisions. I would rather turn down a project up front than deliver something that does not work.
