@sou-hardo
AI researcher and software engineer.
Bio
I'm an engineer, fresh into the industry, though I've been doing research a lot longer than I've been shipping code. Most of my studies went into computer vision for medical imaging, brain tumors, skin lesions, that kind of thing, and I still co-run a research lab on the side because I never quite left.
These days I work on production AI at Ledgercross, and I'm putting a bit more of my focus into building software and trying to build something of my own.
I was the shy kid who moved schools seven times before tenth grade, courtesy of my dad's job. No childhood friends, two rounds of bullying. Character building, allegedly.
Then college happened, where my "never fails an exam" streak died in spectacular fashion. Bombed my board exams too, for good measure, and missed the cutoff for every decent public university in the country.
So I did the only reasonable thing: I deliberately bombed the public university admissions too, on purpose, so my dad would have no choice but to send me to a private university for Computer Science instead. Sorry, Dad.
First semester I woke up at 6am out of pure spite and pulled a 4.00 CGPA. Then I learned a 4.00 unlocked a bigger tuition waiver than a 3.75, so I just kept doing that for four years straight. Ended at a 90% waiver. Capitalism, but make it academic.
Met my best friends Safwan and Jabir on day one by inviting myself to their desk. Years later Safwan admitted they'd clocked me as a nerd and planned to dump all the group work on me. Reader, they were right, I did every bit of it myself, and somehow we're still best friends.
Safwan dragged me into research a semester later. Three journal papers, a stack of conferences, and a self-founded lab after that, I still haven't stopped.
We all finished uni in record time with zero real software projects to show for it, and still turned down unpaid internships on principle. I landed one at Ledgercross first, Jabir and Safwan followed a week later. Instead of frontend busywork, I got to build an AI pipeline to catch deepfakes. Not bad for an intern.
Notes
Notes are coming soon.
Experience
- Co-founded a non-profit, student-led AI research initiative providing mentorship and research opportunities across computer vision, medical AI, and deep learning.
- Now supervise the Computer Vision & Medical-AI track; the lab has grown to 24 peer-reviewed publications spanning medical imaging, explainable AI, and Bengali-language NLP.
- Project Lead coordinating a 5-engineer team building GCP microservices in Go and Triton, scaling an identity/document-verification platform to 10M+ daily verifications with active-active failover.
- Deployed a 5-stage Vision/LLM pipeline automating fraud and biometric checks in under 30 seconds.
- Solo-built a distributed deepfake-verification system: browser extension, orchestrator, and AI analysis agent, backed by Redis, PostgreSQL (pgvector), and S3.
Stats
GitHub activity
Monkeytype
15 seconds
89
100%
30 seconds
80
100%
60 seconds
62
95%
typing time
15h
1,413 tests
Notes
All notesSelected publications
All publicationsT3SSLNet: Triple-method self-supervised learning for enhanced brain tumor classification in MRI
IEEE Access · 2025 · 10 citations
IEEE DataPort · 2024 · 9 citations
KbFL-XAI: Explainable knowledge-based federated learning for eye disease diagnosis
Biomedical Engineering Advances · 2025 · 7 citations