Problem

Different customers required RAG and agent-based AI capabilities, but each had different LLM providers, data types, and retrieval needs. A single fixed RAG implementation would not scale across use cases.

Architecture

• Built a highly configurable RAG and AI-agent architecture adaptable to multiple customer requirements.
• Made core components configurable: LLM provider, embedding model, vector database, chunking strategy, chunk size/overlap, and retrieval settings.
• Implemented document processing, embeddings, vector search, retrieval, and LLM generation pipelines.
• Built agent workflows on top of the retrieval layer for more complex, multi-step tasks.
• Added evaluation and observability tooling to monitor and improve AI application quality.

Outcome

A single reusable architecture that can be configured and deployed for multiple customer use cases, reducing the need to rebuild RAG pipelines from scratch for each new client.

Stack

Problem

The company needed intelligent, real-time fleet monitoring and vehicle-level insights derived from continuously generated vehicle telemetry, to support fleet management and vehicle health decisions.

Architecture

• Data layer: real-time vehicle telemetry ingested via Kafka and processed through an AI/ML pipeline.
• ML layer: anomaly detection and battery health models, including State of Health (SoH) analysis and battery degradation / Remaining Useful Life (RUL) forecasting.
• AI layer: LangGraph orchestration combining telemetry, business rules, and model outputs into coherent, vehicle-level insights.
• Application layer: backend integration delivering insights through fleet-management dashboards.
• Combined rule-based and ML-based detection for abnormal vehicle and driving behaviour.

Outcome

A production-oriented platform combining streaming data, ML models, and AI orchestration to deliver actionable vehicle intelligence — anomaly alerts, safety insights, and battery health forecasts — directly through fleet-management dashboards.

Stack