Favoriot started to implement AI into its IoT platform by using traditional Machine Learning techniques, and many wondered when Favoriot would use Large Language Models (LLMs).
Below are the reasons Favoriot started with Machine Learning before moving to Agentic AI.
1. How is this ML different from LLMs
The ML used in Favoriot today
This is classical and applied machine learning focused on structured IoT data.
Think of:
- Numbers
- Time-series
- Sensor streams
- Device metrics
Its job is to:
- Predict values
- Classify states
- Detect anomalies
- Learn patterns from historical data
Examples:
- Forecast energy usage
- Detect abnormal temperature behaviour
- Classify device health
- Estimate remaining useful life
These models are:
- Small
- Fast
- Domain-specific
- Deterministic in purpose
They learn only from your data, within a defined scope.
What LLMs are
LLMs are a very different creature.
They are:
- Trained on massive text corpora
- Language-first
- Probabilistic in reasoning
- General-purpose
Their strength is:
- Understanding and generating language
- Reasoning across concepts
- Explaining, summarising, planning
They do not natively understand sensor physics, time-series behavior, or device constraints unless wrapped with tools or data adapters.
In short:
| Aspect | Favoriot ML | LLM |
|---|---|---|
| Data type | Structured IoT data | Text, language |
| Scale | Small to medium | Massive |
| Purpose | Prediction, detection | Reasoning, generation |
| Training | On your data | Pre-trained globally |
| Determinism | High | Probabilistic |
They solve different problems.
2. Why Favoriot ML is still essential
A lot of people assume LLMs will replace everything.
They won’t.
For IoT systems:
- LLMs are too heavy for real-time sensor inference
- LLMs do not replace signal modelling
- LLMs cannot forecast time series accurately on their own
- LLMs do not understand noise, drift, or sensor physics by default
Your ML layer is the ground truth intelligence.
It understands reality as numbers behave in the real world.
LLMs sit above this layer, not instead of it.
3. Can this evolve into Agentic AI?
Yes. And this is where things get interesting.
But Agentic AI does not start with LLMs.
It starts with capability loops.
What makes an agent an agent
An agent needs five things:
- Perception
Read data from the environment
→ IoT sensors already do this - Understanding
Interpret patterns and states
→ ML models already do this - Decision logic
Decide what action makes sense
→ Rules + ML outputs already exist - Action
Trigger something
→ Alerts, workflows, and actuators already exist - Memory and learning
Improve over time
→ Scheduled retraining already exists
Seen this way, Favoriot already has 80% of an agent loop.
4. Where LLMs come into Agentic AI
LLMs are not the agent.
They are the reasoning and coordination layer.
An evolution path looks like this:
Stage 1: ML-driven intelligence (today)
- Models predict
- Rules act
- Dashboards explain outcomes
Stage 2: Tool-aware reasoning
- An LLM can:
- Interpret ML results
- Explain why something happened
- Suggest following actions in human language
Example:
“Energy usage will exceed baseline in 2 hours. Recommend load shedding or rescheduling non-critical equipment.”
Stage 3: Goal-driven agents
- An agent is given a goal:
- Reduce energy cost
- Maintain uptime
- Prevent failures
The agent:
- Queries ML models
- Monitors outcomes
- Adjusts actions
- Learns from feedback
Stage 4: Multi-agent systems
- One agent watches energy
- One watches maintenance
- One handles alerts
- They coordinate through shared context
This is where Agentic AI truly appears.
5. Why starting with ML first is the right move
If you begin with LLMs alone:
- You get fluent explanations
- But a weak grounding in reality
If you start with ML first:
- You get reliable signals
- Trustworthy predictions
- Clear cause-and-effect
Agentic AI must be grounded.
Otherwise, it becomes confident but wrong.
IoT is a physical world problem.
Physics beats language every time.
6. The big picture
Think of it like this:
- ML is the sensory and predictive brain
- Rules are reflexes
- LLMs are the thinking and planning layer
- Agents are the orchestration of all three
Favoriot moving into ML is not competing with LLMs.
It is laying the foundation for real-world agentic systems that actually work.
That foundation matters more than hype.
FAVORIOT Intelligence References
- Official Announcement -Favoriot Evolves into an AIoT Platform with Built-in Machine Learning
- Why Favoriot Chooses Machine Learning Over LLMs
- Machine Learning vs Deep Learning in IoT
- Traditional Machine Learning vs Large Language Models (LLMs)
- Why Favoriot’s Built-in Machine Learning Matters for AI Researchers and IoT Developers
- Why Favoriot’s ML Infrastructure Reduces Costs
- AI and ML in IoT, Explained Without the Jargon
- Start Anomaly Detection in IoT with Less Data
- AI and ML in IoT, Explained Without the Jargon
- The Key Differences: Favoriot’s Rule Engine 2.0 and AI Agents
- The Role of Machine Learning in IoT Systems
- Favoriot’s Rule Engine 2.0: A Structured Approach to IoT Automation
- A Quiet Shift Is Coming to the Favoriot Platform
Leave a Reply