Challenges Encountered During Development

Building a production-ready application from the ground up involves navigating a series of technical and strategic hurdles. The WellNest project was no exception, and overcoming these challenges was critical to arriving at the robust, reliable final product.

1. The Python Pathing and Module Resolution Puzzle
   The most persistent technical challenge was a series of ModuleNotFoundError errors. This issue manifested differently depending on how the code was executed, creating a complex problem to diagnose.

The Problem: Scripts like ingest_data.py would fail to import modules (e.g., from db.vector_store), while the Uvicorn server running the main application would fail with the opposite import structure.

Root Cause Analysis: The core issue was the inconsistency of Python's sys.path based on the execution context. Running a script directly (python -m backend.scripts...) adds the current working directory (the project root) to the path, requiring absolute imports like from backend.db.... However, running the Uvicorn server (uvicorn backend.main:app) treats the backend directory as the top-level package, causing those same absolute imports to fail.

Solution: The solution was to establish a single, consistent workflow. We standardized on running all commands from the project's root directory (WellNest/) and refactored every import within the backend to be absolute (e.g., from backend.api..., from backend.services...). This created a predictable environment and resolved all import-related errors permanently.

2. Virtual Environment (venv) Management and Dependency Conflicts
   As the project grew, a subtle but critical environment issue emerged, leading to commands like uvicorn suddenly not being recognized.

The Problem: A new, secondary virtual environment (.venv) was accidentally created in the project root, while all the project dependencies had been installed in the original environment (backend/venv).

Root Cause Analysis: The terminal was automatically activating the new, "empty" virtual environment, which lacked all the necessary packages. This highlighted the importance of a clean and singular source of dependencies.

Solution: We diagnosed the issue by inspecting the installed packages in both environments using pip freeze. We then consolidated our workflow by deleting the extraneous environment and establishing a strict procedure: always activate the correct environment (backend/venv/Scripts/activate) from the root directory before running any commands. This ensured that we were always working with the correct set of tools.

3. The Data Scarcity and Quality Problem
   The most significant strategic challenge was the lack of a suitable public dataset for a mental wellness chatbot.

The Problem: Most publicly available data consists of unstructured, "noisy" forum discussions or chat logs, which are filled with anecdotal advice, conflicting information, and are fundamentally unsafe to use for a support tool.

Root Cause Analysis: A reliable AI must be grounded in reliable data. Using low-quality data would directly lead to a low-quality, unpredictable, and potentially harmful user experience.

Solution: We pivoted from data sourcing to data engineering. A custom dataset was meticulously created from factual, vetted sources and structured into a highly-optimized Question & Answer format within several PDF documents. This strategic decision was the cornerstone of the project's success, transforming the RAG pipeline from a generic tool into a precise, safe, and highly effective support system by giving the retrieval algorithm perfect targets to find.