Raj Aryan

Context-aware mental health support assistant ('Mimi') powered by custom RAG pipelines and a hybrid database architecture.

problem

Mental health chatbots frequently suffer from hallucinations or ungrounded medical advice, while existing architectures struggle to balance strict transactional compliance with unstructured chat logs.

architecture

• Custom RAG Pipeline
• Pinecone Vector Store
• Ollama Local LLM
• MySQL User ACID transactions
• MongoDB Chat history logs

decision

Enforced strict context constraints using verified psychological reference datasets; isolated relational user profile state from non-relational chat streams.

result

Top 100 finish out of 500+ teams; 100+ concurrent session throughput with zero medical hallucinations.

Secure document analysis pipeline leveraging local compute tunneling and TiDB vector search for zero-cost document intelligence.

problem

Cloud-hosted LLM APIs incur steep per-token costs and impose memory constraints on free/low-tier instances when parsing multi-page technical documents.

architecture

• AWS Lightsail
• Local Tunneling Ollama Inference
• PyMuPDF
• TiDB Vector Embeddings
• Cloudinary File Storage

decision

Decoupled cloud frontend/storage from compute-intensive LLM inference via secure tunneling, driving external API cost down to $0.

result

50+ active signups; processed 100+ multi-page PDFs with zero external API fees.