RAG chatbot assistant

GulzarSoft AI Assistant

A custom React Native AI chatbot that answers questions about GulzarSoft services, projects, pricing, contact details, and expertise using structured retrieval.

QueryRAGRankAnswer

Company knowledgeClient assistantContext-aware replies

Case study

Problem, role, and solution.

A quick read of what had to change, where the engineering ownership sat, and how the final system answered the core project need.

Problem

GulzarSoft needed a fast client-facing assistant that could answer service, pricing, portfolio, contact, and project questions accurately without relying on a static FAQ or generic chatbot responses.

My role

Designed the retrieval architecture, FastAPI backend, LLM orchestration, ranking logic, confidence handling, intent classification, and React Native chat experience.

Solution

Built a retrieval-powered chatbot that classifies intent, detects negation, expands queries, retrieves relevant company knowledge, reranks results, and uses multi-document context to generate grounded responses.

System evidence

Real screens from the GulzarSoft AI Assistant workflow.

These supporting visuals show the practical workflow, implementation details, and output quality behind the project.

GulzarSoft AI Assistant screenshot showing the mobile chatbot answering company and service questions. — Assistant conversation flow

GulzarSoft AI Assistant screenshot showing retrieved service context and a structured assistant response. — Grounded service retrieval

GulzarSoft AI Assistant screenshot showing project and portfolio question handling in the mobile app. — Project discovery answers

Technical profile

Stack, integrations, and build risks.

A compact read of the tools, connection points, and engineering constraints behind this case study.

12 tools

Technology stack

React NativeFastAPIGroqLLaMA 3.3 70BLangChainMultiQueryRetrieverHuggingFace EmbeddingsChromaDBTF-IDFCosine similarityIntent classificationConfidence scoring

6 links

Integrations

Groq APIGulzarSoft service knowledge baseGulzarSoft portfolio contentChroma vector databaseFastAPI chat APIReact Native mobile UI

5 risks

Engineering challenges

Handling vague client questions with useful clarifying behavior.Ranking retrieved documents so the assistant returns relevant company-specific answers.Detecting negation and intent shifts in short conversational prompts.Balancing fast response time with multi-query retrieval and reranking.Keeping generated answers grounded in available service, pricing, project, and contact context.

Outcomes

What changed after delivery.

4 verified results

Delivered a mobile AI assistant for company service and project discovery.

Enabled accurate answers about services, pricing, portfolio, contact details, and project insights.

Improved retrieval quality with query expansion, vector search, TF-IDF, cosine reranking, and confidence scoring.

Created a reusable GenAI pattern for client interaction and structured knowledge access.