Mindteck
Randomwalk
Agentic Unified Testing Platform
--:--:--
— ms
CONNECTING…
?
production environment · last sync just now · nodes: —
sessions analyzed · noise suppressed · knowledge base entries
1 2 3 4 5
◆ LOGMIND
Upload CTC logs · Noise suppression · AI root-cause hypotheses · Human-in-the-loop verdicts
CONNECTING
◆ ANNOTATION ENGINE
Verdict history · Admin notes · Correction loop to MT RootHub
CONNECTING
◆ KNOWLEDGE BASE
Confirmed resolutions · Learns from engineers · Feeds AI context on every run
CONNECTING
◆ INFERENCE HUB
Parameter dictionary · Value ranges · Injected into every AI analysis run
CONNECTING
◆ TRAINHUB
Periodic LoRA fine-tuning from MT RootHub · Promotes improved model weights
SCHEDULED
Recent sessions
No sessions yet
Latest resolutions
No resolutions yet
◆ LOGMIND
Upload Session Logs

Drop CSV files here or browse

CTC log CSV · Multiple sessions supported
Activity Stream 0
LOGMIND
No session loaded. Drop a CTC log file on the left to begin.
Events Over Session
Rows by Module
Error Mix
◆ ANNOTATION ENGINE
Verdict Mix
Add Resolution Manually
Loading…
◆ KNOWLEDGE BASE
Search
Activity Stream 0
Confirmed resolutions feed into the AI's context on every analysis run. Read-only — entries are added via MT Annotation Engine.
Loading…
◆ TRAINHUB
Engineer-authored resolutions are distilled into the model via LoRA fine-tuning. Training starts on its own once enough new corrections accumulate — there is no schedule and no start button.
Training Pairs
Engineer-authored
New Since Last Run
Counts toward trigger
Already Trained
Baked into weights
Status
Auto-triggered
Auto-Training Trigger
— / —
Loading…
Pipeline Status
Knowledge base exported
Awaiting threshold…
Fine-tune job prepared
Training completed
Awaiting ignition…
Adapter promoted to production
Awaiting deployment…
How It Works

1. Collection
Engineers correct AI hypotheses in MT LogMind. Each correction lands in MT RootHub keyed by its fault signature.

2. Eligibility
Only resolutions an engineer actually wrote are used. Hypotheses merely confirmed as-is are retrievable, but never train the model — fine-tuning on its own output teaches it nothing.

3. Trigger
When enough new engineer-authored pairs exist, a run starts automatically.

4. Promotion
The trained adapter is registered with Ollama so the next analysis uses the updated weights.

Training Dataset Preview
◆ INFERENCE HUB
Dictionary of every CTC parameter — symbol, full name, unit, expected range. Automatically injected into the AI prompt on every analysis run.
Live context injection: entries here are passed to the AI on every analysis run.
Symbol
Full Name / Description
Unit
Normal Range
Min → Max
Actions
Loading schema…
Add Parameter