Building Full GenAI Applications

Trace a document question-answering request through the user interface, application logic, model provider, and returned response.

Create a Python project structure with reproducible dependencies, environment-based configuration, and protected API credentials.

Send an equivalent chat-completion request through the Mistral API.

Implement a common model interface that switches among Groq, Mistral, and Ollama through configuration.

Construct reusable system and user prompts that constrain a model's role and output format.

Extract clean text and source metadata from uploaded PDF and plain-text documents.

Split extracted text into retrieval chunks with an appropriate size and overlap.

Generate embeddings for text chunks and use cosine similarity to compare semantic relevance.

Store document embeddings and metadata in a persistent vector index.

Retrieve the top relevant chunks for a user question and inspect their similarity scores.

Construct a grounded generation prompt from the question and retrieved context.

Attach verifiable source citations to an answer using retrieved document metadata.

Return an explicit insufficient-evidence response when retrieved context cannot support an answer.

Implement a FastAPI document-ingestion endpoint that validates a file and builds its vector index.

Implement a FastAPI question-answering endpoint that runs retrieval and grounded generation.

Build a Streamlit interface for document upload, model selection, and conversational questions.

Connect the Streamlit interface to the FastAPI endpoints and display streamed answers with citations.

Maintain separate document indexes and conversation histories for independent user sessions.

Handle provider timeouts, rate limits, unavailable local models, and malformed responses without crashing the application.

Protect secrets and enforce file-type, file-size, and input-length limits.

Record structured logs for request latency, selected provider, retrieval results, and model usage.

Create a small evaluation set containing answerable, ambiguous, and unanswerable questions for a sample document.

Measure whether retrieval returns the source passages needed to answer each evaluation question.

Assess generated answers for groundedness, citation correctness, and appropriate refusal when evidence is missing.

Compare Groq, Mistral, and Ollama results by answer quality, latency, privacy, and estimated cost.

Tune chunk size, chunk overlap, retrieval count, and prompt instructions in response to observed failures.

Rebuild the core retrieval-and-generation workflow with LangChain and map each abstraction to the corresponding raw-Python component.

Write automated tests for document chunking, retrieval, provider switching, and the question-answering API.

Containerize the FastAPI and Streamlit services and run the complete application in hosted-model mode.