AI where it earns its place. And nowhere else.
We use AI in three places: watching production, in the agents we build for clients, and in our own products. Each time it is applied narrowly and on purpose, because a model nobody can interrogate is a worse colleague than a rule everybody understands.
- In operations
- Anomaly detection, alert correlation and forecasting
- For clients
- AI agents on WhatsApp, documents and meetings
- Our own products
- MeetNoTime · AskMe · DaeForge (in development)
- Guardrail
- Reviewed, logged and reversible — never unsupervised
What sits behind the claim
Concrete uses, each one scoped to where a rule or a person would do worse.
- Anomaly detection. Baselines learned from your own traffic, so alerting fires on genuine deviation in signals with daily and weekly shape, where a fixed threshold cannot work.
- Alert correlation. Related signals grouped so an incident arrives as one event with context, rather than forty alerts from forty components describing the same outage.
- Forecasting. Disks and connection pools that will run out in days, and latency drifting a little each release, caught as a trend before they become an outage.
- Incident context. The likely blast radius and the recent changes to a service surfaced alongside the alert, because the honest first question in any incident is what changed.
- AI agents for clients. Assistants that answer customers on WhatsApp, read documents and summarise meetings, built into the business systems they serve.
- Meetings and documents. Transcription, summaries, action items and notes in English, Hindi or Hinglish with live translation in MeetNoTime.
How it works in an engagement
Measurement first, then automation you can review, on a stack you own.
A baseline before a model
We instrument what you already run and establish a baseline first, because anomaly detection on an unmeasured system is theatre. The measure of success is fewer pages, not more detections.
Automation you can review
The responses your team already performs from a runbook — restarting a known-bad process, expanding a volume, draining a node — become version-controlled automation, reviewed like code, logged when it fires and reversible. It is not a licence to let a model act unsupervised.
Agents built like any other system
AI features are engineered with the same pipeline, tests and security as the rest of the platform. Python carries the document processing and the model-serving layer behind them, and the agent connects to the systems your business already runs.
Our own products first
We build and run AI products ourselves: MeetNoTime is live, AskMe answers questions from your databases, spreadsheets and CRM, and DaeForge, our own assistant that keeps tasks, notes and decisions in one place, is in development.
A model nobody can interrogate is a worse on-call companion than a threshold everybody understands.
So AI sits behind anomaly detection, correlation and forecasting, and nowhere else in operations. Everything runs on Prometheus, Grafana, Loki and the ELK stack rather than a proprietary platform, so you own the data. The full service is on the AIOps page.
Questions we are asked
What is AIOps, and how is it different from monitoring?
Monitoring collects and displays signals. AIOps acts on them — correlating related alerts into one incident, learning normal behaviour per service instead of applying one static threshold, and automating the responses that are already well understood. Monitoring is the prerequisite, not the same thing.
How does anomaly detection avoid becoming more noise?
By being scoped to where it earns its place. It is applied to signals with genuine seasonality, where static thresholds cannot work, and every detection goes through the same correlation layer as everything else.
What does operations automation actually automate?
The responses your team already performs from a runbook: restarting a known-bad process, expanding a volume, draining a node, clearing a stuck queue, scaling for a predictable peak. Each is version-controlled, reviewed like code, logged when it fires and reversible.