medium Post: Journey Through a Recent AI Project

mapping-the-parallels.md

Project Stage	Project Storyline	Hiking Analogy Storyline
Project Initiation	Started a machine vision and Gen-AI solution for an e-commerce company, expecting a straightforward implementation.	Embarked on a hike to a renowned Grenadian waterfall, expecting an easy journey based on initial guidance.
	Embarking on a journey to develop a machine vision and Gen-AI based solution for auto-tagging and recommending related products for an e-commerce company, I was as optimistic as a traveler setting off on a hike to a renowned Grenadian waterfall, promised by the initial project outline to be a straightforward endeavor. Much like the brochure that assured a hike suitable for all ages, the initial system design and solution architecture of our project seemed easily achievable, suggesting a smooth path ahead.
MVP Development	Quickly implemented the MVP, leveraging GPT for streamlined processing with accurate outputs.	The early stages of the hike were easy, with a well-defined path and scenic views.
	The implementation of the Minimum Viable Product (MVP) was swift, akin to the early stages of the hike, with the heavy lifting gracefully handled by GPT, ensuring the end-to-end processing flow was as streamlined as the path leading to the waterfall. The outputs, much like the scenic views along the hike, were accurate and of high quality, fostering a sense of accomplishment and anticipation for the journey ahead.
Scaling Challenges	Encountered difficulties in scaling, achieving sub-second response times, and managing costs with GPT APIs.	Midway through the hike, the path became unexpectedly tough, making progress more difficult.
	However, as we ventured further into the project, akin to reaching the midpoint of the hike, the challenges of scaling the solution, ensuring sub-second response times, and controlling the costs associated with using GPT multi-modal APIs became apparent. The goal, much like the waterfall, seemed distant, and the path ahead, more daunting than initially promised. It was at this juncture that, much like the guide's humorous reassurance in the face of exhaustion, a shift in perspective was needed to lighten the spirits and reinvigorate the project.
Strategy Shift	Pivoted to using foundational models (CLIP, DINO) and a Vector Database, precomputing embeddings for efficient retrieval.	Found a clearer, more manageable path to navigate the challenging terrain.
	Re-architecting the processing pipeline to incorporate more primitive models, such as CLIP and DINO, alongside a Vector DB, was akin to adjusting our stride and finding a new path through the terrain. By pre-computing all embeddings and object identification at indexing time, and utilizing simple text and image patch vector search queries over the pre-computed index, the project found its rhythm, achieving sub-millisecond query responses. This approach, much like finding humor in the guide's words, transformed the journey, making the once daunting task of scaling the solution not only achievable but trivial, with costs dramatically reduced and complexity managed more efficiently.
Optimized Architecture	Achieved sub-millisecond query responses, reduced costs, and built a robust, scalable solution.	Adjusted pace and approach, making the final stretch of the hike easier.
	The final solution architecture, clean and transparent, stood as a testament to the journey undertaken, much like the breathtaking view of the waterfall that awaited us at the end of our hike. The increase in the complexity of the pipeline processing, initially a concern, was dramatically reduced, mirroring the unexpected ease found in the latter half of the hike, especially with the adoption of a free open-source vector image search framework to manage the workloads.
Final Outcome & Reflection	Delivered a clean, efficient architecture, learned valuable lessons in adaptability, and created a strong foundation for future growth. 🚀	Reached the waterfall, reflecting on the journey’s challenges and lessons in perseverance and flexibility.
	In reflection, the journey to develop this machine vision and Gen-AI based solution, much like the hike to the Grenadian waterfall, taught invaluable lessons in perseverance, the importance of humor in the face of challenges, and the openness to adapt and embrace the unexpected. The end solution, solid and extendable, is not just a product for the client but a milestone in a journey filled with learning and growth.