Repetitive support
The same questions every day tying up your team. An agent answers, qualifies and hands over to a person only what needs a person.
AI applied to the business
AI integration consulting for companies
I put AI inside your system or operation where it cuts cost or removes a bottleneck: LLM agents, intelligent search over your own data, process automation and voice. From proof of concept to production, with cost and quality under control.
- Free initial conversation
- AI in production: agents, voice and text classification
- Direct contact with the person who designs and codes
Jonathan Bispo · Software Architect · 15+ years · MSc (UFMG)
When it makes sense
Where AI usually pays off
I don't sell AI as a trend: I start from the problem and only continue if the return is real.
Information nobody can find
Documents, spreadsheets and databases scattered around. Natural-language search over your own data (RAG), showing the source of each answer.
Manual work with text
Classifying, summarising and extracting data from emails, forms and open-ended answers, instead of typing and mass reading.
Processes that need people to connect the dots
Automation across systems: the AI interprets and decides where things go, the system executes and everything is logged.
Phone support and research
Voice with real-time transcription and synthesis, for triage, interviews and outbound contact.
A system that already exists
Add AI to what you already have, through integration, without rewriting everything, and with a person in the loop when the decision matters.
What I deliver
From assessment to a working system
Every stage has a clear deliverable, so you can decide whether to continue.
Assessment and prioritisation
I map where AI really pays off, with risks, estimated cost and what to do first.
Proof of concept with real data
A working prototype on your own data, to validate quality and feasibility before investing in the whole.
Integration in production
Agent, search or automation connected to your system through APIs, with authentication, queues, logs and monitoring.
Cost, quality and security under control
Spending limits, answer evaluation, handling of sensitive data and data protection compliance.
Documentation and handover
Documentation and technical sessions with your team, so you don't depend on me to operate it.
Continuous improvement
Follow-up after go-live: prompt and model tuning, cost control and new use cases.
How it works
A short path to production
You see the result of each stage before deciding on the next.
-
1
Initial conversation
30 to 60 minutes, free of charge, to understand the problem, the data and what has already been tried.
-
2
Assessment and proposal
You get the solution design, scope, risks and investment, in plain language.
-
3
Proof of concept
A prototype on real data, with success criteria agreed beforehand.
-
4
Production and evolution
Integration into your system, monitoring and adjustments based on real usage.
Experience
AI that already runs in production
Real projects, with real users, running every day.
Voice AI · Telephony
AI voice interviewer over a phone call
A real-time pipeline of transcription, LLM understanding and speech synthesis on top of Asterisk, running phone interviews with no human operator.
AI agents · Automation
Agents embedded in daily operations
An MCP server authenticated against the product's database, exposing safe tools scoped to each user, plus agents that monitor quotas, trigger alerts and run routines.
Machine Learning · NLP
Automatic classification of open-ended answers
BERT fine-tuning in PyTorch to classify open-ended answers, preserving the context across questions from the same respondent.
Frequently asked questions
Before you start
How much does it cost to integrate AI into my company?
It depends on the scope, so I start with a free conversation and assessment. The proposal states the investment before any work begins, in whatever format makes sense: a fixed-scope project, monthly hours or a one-off assessment. I also estimate the running cost (model usage and infrastructure), which has to be part of the maths from the start.
Is my data safe? Do I have to send it to third parties?
I assess this case by case. Depending on how sensitive the data is, the options range from using providers with contractual privacy guarantees and masking information before it is sent, to running models on your own infrastructure. Data protection law is part of the design from the beginning.
Do I need my data organised before starting?
Not necessarily. Assessing the quality and organisation of your data is part of the discovery, and often the first step of the project is to tidy up the minimum needed for the use case to work.
Will AI replace my team?
The goal is to take repetitive work off your team, not the decisions that matter. In sensitive cases I design the flow with a person reviewing or approving before the action happens.
Which AI model or provider do you use?
I choose based on the problem, cost and privacy, not brand. It can be the API of a major provider, an open model hosted by you, or a combination. I aim to keep the architecture decoupled, so switching models later doesn't mean rebuilding the system.
Do you work with systems that already exist?
Yes, it is one of the most common cases: I integrate AI into your current system through APIs, without rewriting it. Beforehand, I run a technical assessment of what exists to see where the integration fits.
Shall we see where AI makes sense in your business?
Tell me in a few lines what you need. I answer personally and, if AI isn't the best answer, I'll say so.