AI agent
development.
An AI agent is software that uses a large language model to decide which steps and tools to use to complete a task — for example looking up data, calling an API and drafting a reply. I build agents with tool calling, retrieval (RAG), memory, logging and human approval for high-risk actions.
I build AI agents that complete real tasks inside your business — looking things up, calling your APIs, drafting responses and updating records — with guardrails, logs and human approval where the stakes are high.
What I build
Tool-calling agents
Agents that call your APIs, databases and third-party services with typed inputs.
Multi-agent workflows
Specialised agents for research, drafting and review, coordinated by a planner.
RAG and memory
Agents grounded in your documents, with short- and long-term memory.
Human approval
Approval queues before an agent sends, pays or deletes anything.
Problems I solve
- Agents that loop, hallucinate tools or take unsafe actions
- No record of why an agent did something
- Vendor lock-in to one model provider
Technologies I use
- OpenAI
- Claude
- Gemini
- Grok
- DeepSeek
- Ollama
- Structured outputs
- Python
- FastAPI
- PostgreSQL + pgvector
- Redis
Development process
- 01
Discovery
- 02
Architecture
- 03
Build
- 04
Test
- 05
Deploy
- 06
Iterate
Relevant projects
Technical approach
- →OpenAI and Claude handle complex reasoning and tool calling where accuracy matters most.
- →Gemini is used for long documents and multimodal inputs such as images and PDFs.
- →Grok and DeepSeek are options for cost-sensitive or high-volume steps, chosen by measured quality.
- →Ollama runs open models locally or on your servers when data must not leave your infrastructure.
- →An LLM router picks the model per step, so switching providers is a config change, not a rewrite.
- →Every tool call is logged with inputs and outputs, and risky actions pause for human approval.
Frequently asked questions
Can an agent act on its own?
For low-risk actions, yes. For anything involving money, customers or deletion, I add an approval step.
Which models do you support?
OpenAI, Claude, Gemini, Grok, DeepSeek and local models via Ollama, behind one routing layer.
How long does a typical project take?
It depends on scope. A focused feature or fix can take days; a first production version of a product usually takes several weeks. I give a written estimate after a short discovery call.
Related services
RAG Development
Retrieval-augmented generation that answers from your own data — ingestion, chunking, vect…
AI Development
AI integrated into real products — not chatbot widgets. LLM features, document intelligenc…
AI SaaS Development
AI-native SaaS platforms with authentication, subscriptions, dashboards, usage limits and …
Python Backend Development
Scalable backend systems in Python — clean APIs, background jobs, caching and integrations…
Ready to start
your project?
I work remotely with startups and businesses across the United States and internationally. Tell me what you're building and I'll reply with next steps.