Marwin Realtime Chatbot
Chatbots
Data Science
Gov
logistics
Summary
The Marwin Team often works with modern radar systems, where documentation can be complex and difficult to navigate—even for technical specialists. Finding the right information quickly is a constant challenge.
To simplify this process, the Postdata Team built a secure internal chatbot tailored specifically for radar and aviation operations. Unlike generic assistants, this AI chatbot is trained on mathematical and physics-based content, allowing it to handle highly specialized queries. With full bilingual support in English and Thai, it delivers precise, context-aware answers directly from documentation and operational data.
Project overview
Our objective was to build a chatbot that could serve as a trusted co-pilot for technical teams. To achieve this, we integrated:
Domain-specific search on math and physics formulas to support radar and aircraft operations.
A RAG pipeline that combines Google Gemini (LLM) with Pinecone vector search for precise answers grounded in verified documentation.
Language detection logic that ensures responses match the user’s input language — English or Thai — without mixing the two.
Firestore persistence for chat history, making the assistant aware of prior discussions while maintaining security and user-level access controls.
WebSocket real-time communication for instant interaction, crucial in time-sensitive operational settings.
We designed the system as a modern cloud-native solution, ensuring scalability and reliability even under demanding conditions.
Results
With Marwin, we proved how a focused AI system, built with modern tools and tailored objectives, can evolve into a trusted assistant for radar and aviation operations. By combining domain-specific knowledge with real-time, bilingual interaction, we delivered tangible improvements and laid the groundwork for broader applications.
Cutting-edge technology stack: Built with FastAPI, React, Firestore, Pub/Sub, Gemini, and Pinecone, the solution shows how state-of-the-art technologies can be orchestrated into a production-ready system.
Operational efficiency: Our engineers and operators can instantly retrieve physics formulas, technical definitions, and radar-related documentation without manual searching.
Bilingual accessibility: The assistant responds fully in Thai or English, depending on the user’s query, enabling smooth multinational collaboration.
Contextual intelligence: By storing prior conversations, Marwin keeps context across sessions, making technical interactions fluid and consistent.
Insights & Conclusions
Project duration:
4 weeks
Team
2
Full-Stack Developer, ML Engineer
Technologies
FastAPI, React, Firestore, Google Pub/Sub, Google Cloud Storage, Pinecone, Gemini (Google Generative AI), LangChain, Docker + Google Cloud Rune
Tech challenge
Bilingual logic: Designing separate prompt templates for Thai and English, ensuring natural translation and domain accuracy without code-switching.
Domain alignment: Incorporating physics and math formulas into embeddings so the chatbot could reason over them correctly.
Latency management: Keeping responses fast despite multiple moving parts (Firestore, Pinecone, Gemini, Pub/Sub).
Secure access: Documentation about the radars contains a very sensitive data, leakage of each could cost a huge amount of money. We provide a secure connections, which keep all the information about the communication with the chatbot private.
Solution
We combined the strengths of retrieval-augmented AI, bilingual support, and cloud-native engineering to deliver a chatbot purpose-built for aviation and radar operations. By grounding responses in official documentation through Pinecone and Gemini, our chatbot avoids hallucinations and provides trustworthy, verifiable answers that technical teams can rely on.
