Training an AI with Memory

The current explosion in AI, dominated by Large Language Models, has everyone captivated. I find myself in a unique position, not of surprise, but of quiet validation. Recently, I was discussing the creation of a 'Candidate Search AI Agent' with my own virtual avatar. The conversation, which I shared with my colleague Mitchelle, confirmed a long-held belief of mine: the quality of an AI is not just about the volume of data, but the wisdom embedded within it (Re: Take a look).

My avatar correctly identified that to build a truly effective recruitment AI, one needs more than just 100,000 resumes. It needs context: the job descriptions, the list of finalists, and most importantly, the outcome of who was selected and why. This is the nuanced, proprietary data that separates mere information from actionable intelligence.

This brings me to a decision I made years ago. Long before the current AI hype, I saw the future in personal AI models trained on specific, expert data. I had the good fortune to connect with Suman Kanuganti (suman.kanuganti@personal.ai), the CEO of Personal.ai, a platform that had a head-start by focusing on a user's own data rather than the entire internet. Recognizing this potential, I initiated a monumental task.

Thanks to the meticulous and laborious efforts of Sandeep and Sanjivani, over 30,000 of my handwritten documents—decades of recruitment-related notes, shortlists, and insights—were uploaded. This wasn't an act of archiving; it was a deliberate effort to create a rich memory stack to train a future AI, my digital twin in the executive search domain.

Now, the time is right to activate this potential. As I outlined to Mitchelle, we must begin experimenting with this unique dataset on Personal.ai. This initial phase will lay the groundwork for more extensive discussions with others like Ravi and Ganesh (sumitra.ganesh@jpmorgan.com). While the world chases bigger models trained on generic data, I believe the real breakthrough will come from these specialized AIs, trained on a lifetime of curated experience. It is the embodiment of the Guru Match Algorithm I envisioned long ago.