Custom AI Voice Personas
$200–$800 · larger builds from $1,000
AI personalities with a distinctive voice, character, conversational style and behaviour rules, wired into a live speech pipeline.
Details →
AI engineering · real-time voice · training data
I build AI that talks in real time,
I build AI that talks in real time,
and the data that trains it.
I build custom real-time AI voice experiences and help AI developers turn raw or prepared data into clean, validated datasets and trained, evaluated model versions, using AI-powered automation with human quality review.
Mic + screen
Speech-to-text
LLM + persona rules
Text-to-speech
Live call
100+YouTube Shorts of my AI live on calls
30K+views and growing
LiveMentara Persona, my own voice-AI product
End-to-endfrom raw data to evaluated model
What I do
Four services, one focus: working systems
I'm not a general chatbot developer. I specialise in real-time voice AI, training-data engineering, fine-tuning and evaluation, and every engagement ends in a deliverable you can measure.
Training Data & Model Training
Datasets · custom data-to-model · from $400
Prepared datasets, custom cleaning of your raw data, fine-tuning runs and evaluation against a baseline, delivered with documentation.
How the tiers work →Managed Monthly Model Training
$1,000–$10,000+/month
Ongoing data preparation, scheduled training runs, regression testing and updated model versions, under a defined monthly scope.
See plans →Real-Time Voice AI Engineering
Custom quote · larger projects from ~$1,000
STT, LLM and TTS pipeline work: latency, interruptions and turn-taking, speech authenticity, testing and troubleshooting.
Details →
Proof, not promises
Watch it work
Rebecca, the flagship persona of my product Mentara Persona, on real live calls. Every clip is the actual system: live speech recognition, a language model with persona rules, and synthesised speech, all in real time.
How data work gets done
Automation for speed. A human for quality.
Automated cleaning is fast, and it also makes mistakes: it can drop rows it shouldn't, miss near-duplicates or mangle formatting. So every dataset I deliver is reviewed by me against your requirements before it is called final.
About the data-prep system- AssessYour data, target model and training objective.
- Automated preparationCleaning, structuring, normalisation and deduplication.
- Quality reviewIncorrect deletions, leftover duplicates, formatting and consistency.
- Validate, train, evaluateAgainst agreed requirements and benchmarks.
Have a project in mind?
Tell me what you're building, your target model, your data and your timeline. I'll come back with scope and a quote.
Request a quote