What I Learned Working with a Giant ‘Evil' Tech Company — Part 1 (or: How I Helped Train an A.I. Model Without Losing My Soul)
You know which one.
From May 2023 to July 2024, I worked on a long-term A.I. content evaluation project through Tech Mahindra (agency) for one of the largest, most powerful tech companies on the planet.
Now, I can’t name them (NDA), but...they might rhyme with "Doodle."
The project focused on assessing generative A.I. performance at scale—used by millions of users worldwide.
What started as basic annotation turned into a masterclass in managing scale, building documentation, designing training systems, and evaluating human-A.I. interaction.
Peep the Comments for More 👇☝🏾
3
3
504
Hope everyone had a great holiday! 😊
Quick update on a project I am building called The Archives.
The project focuses on whistleblowers, researchers, and historical figures connected to alleged advanced aerospace programs and unconventional technology.
As the project grew, manually adding people and research via prompt became inefficient, so I prompted a lightweight CMS to better organize and scale the archive. I was blown away at how easy it was to do with Lovable (https://www.linkedin.com/company/lovable-dev/) and some Supabase (https://www.linkedin.com/company/supabase/) database integration.
Stay tuned for more updates!
15
275
Fundory — Grant Intelligence for Nonprofits
Grant funding runs on a brutal arithmetic: most applications fail, and the ones that fail still cost weeks of a small team's time. Nonprofits with two or three staff are spending that time writing into opportunities they were never positioned to win.
Fundory attacks the problem upstream. Instead of helping organizations write faster, it decides what's worth writing at all. Every discovered grant is screened against the organization's actual profile and returned with a fit score, so a 65% never consumes the effort a 91% deserves. The product states its threshold plainly: apply above 85%.
The pipeline runs Discovered → Screened → Waiting for Approval → Drafted → Closed, with AI moving work forward and humans holding the approval gate. Drafted proposals carry two separate scores — model confidence and draft quality — because those measure different risks and collapsing them would hide the one that matters.
I built the screening and scoring layer, the document and narrative system that grounds drafts in the organization's real history, and the pipeline that keeps 28 live opportunities legible to a team without a grants department.
Honest signal over generated volume.
A singer songwriter for a music video her name is VOX 808
0
67
My first ever hackathon
2
158
Client was a Managed Services company that provided IT
Infrastructure to mid-sized clients. This was a series of automations that acted as Level 1 support. The first n8n workflow monitored their support email inbox for incomming service requests. If the request did not have all the required information it triggered an outbound call from our AI assistant that would interview the customer to gather all the requirements before
completing the service request. Password resets, create and manage O365 Distribution Groups and shared mailboxes, etc...
0
145
I created and published Mocky, an interview/communication coach, on the App Store by myself. Tools used included Replit, Claude, and ElevenLabs. https://mocky-app.com/