Summary
Highly skilled AI Software Engineer with hands-on experience in developing advanced language and machine learning models. Specialized in designing frameworks and solutions for PDF manipulation, legal automation, and model-based detection of hateful content. Passionate about pushing the boundaries of AI applications, from coding support to next-generation PDF frameworks.
So Amazon’s Scientist or Senior for AGI, asks you to help fine tune models, probably those that run on AWS (after u train it and save it u have a model) (then u could fine tune it and save it again; then u license AWS to run ur file; but to run it u need to provide them the inference code as well) However I am lacking front end proprietary UI skills of using AWS 😭 That's why, after they hit the 15% H1B threshold at only 3000 approved visas per year? They're required to conduct thorough outreach to made sure they cannot find a qualified American to flll the role 😭 for each 1 new visa applied for (probably 20-30k) out of the extra 7000 they get approved 👌 Sadly fine tuning an LLM is always the opposite direction from AGI 👏
Portfolio
Core Skills
- Proficiency in AI Model Development (LLM, BERT-based models)
- Expertise in PDF manipulation frameworks (C++ TFLite)
- Familiarity with coding assistants for exploit generation and advanced Googlebot usage
- Experience with web technologies (Next.js, React)
- Strong background in AGI and multi-model integrations
Experience & Projects
- SageEmail: Say goodbye to manually labeling, organizing, and backing up your emails (and later any other type of data) forever! SageEmail brings you an unbeatable (literally, by far) base advanced AI model to auto-sort your emails into routine alerts, important alerts, important non-alerts, and then spam/marketing. SageEmail automatically backs up all important emails, as well as alert you via Push Notification or SMS etc. for all high important alerts. (User adjustable settings) Users will also be able to custom define labels, select training data for that label or for anything else, and implement advanced AI automation, then easily retrain or fine-tune on SageEmail Kuberentes GPU instances, controlled by SageEmail's intuitive web-UI frontend! Initially SageEmail will be availble for Gmail only by using Google SSO an the Gmail API, and please expect an inital product release by 1 week from now or around mid-March 2025!
- Code LLM: Developed and trained a LLM specialized for coding tasks, improving accuracy and developer productivity.
- Logic LLM: Developed and is perfecting the smallest possible highest-relative ability model there is to produce logically useful generation results! This model may not only come in handy when a user needs to run a tiny model locally or locally on edge, but it represents the highest tier research that’s accessible to normal people who want to learn how frontier and advanced LLM engineering and research techniques are done.
- Safety LLM: Created a general knowledge LLM designed for users to practice LLM design (it'll be open source on Github with instructions and insights when polished) as well as the ability for users to practice training safety stop tokens into the LLM, and to implement HFRL for both performance and safety.
- Exploit Googlebot: Built a gym-based PriorityBFS Googlebot aimed at sophisticated exploitation proof-of-concepts and penetration testing.
- Exploit LLM: Developed an LLM for generating exploit code and security insights, elevating the platform’s security research capabilities.
- StupidityModel (HateCrimeModel): Implemented a BERT base uncased model to intelligently detect hateful content.
- PersonalModel: Created a Bert-base-uncased model as a Chrome addon, customizing AI interactions to individual user preferences.
- SageFlower: Engineered a 1024x1024 flower generation model under GPU constraints, demonstrating optimized image synthesis.
- SagePDFFramework: Assisted in building a C++ TFLite-powered PDF manipulation framework, aimed to surpass existing solutions in performance and features.
- SageLegal: Worked on a complex multi-model AGI program, enabling near-automated legal form handling and advanced documentation analysis.
- TwitchModel: Created a foolproof bert-base-uncased model, that supports easy personalization, fine tuning, and retraining, that successfully classifies toxic+negative comments on Twitch, and also separately Spam comments.
- TwitchSpamModel: A highly effective sub-part of the TwitchModel, extremely good at detecting spam messages and optionally hiding them.
Education
- CS50 dropout, Harvard
- Bachelor of Science in Engineering Science, Tufts
Experience
- AI-enhanced software engineer, PDFSage Inc. Jan 2025 - Present
- Non-AI-enhanced entrepreneur, less successful companies, 2021 - 2024
- Product Manager, Kensho Technologies. 2013-2014
References
Available upon request.
Work Gaps
I do have some work gaps, but please refer to my portfolio! There’s no bullshitting here, and if you’re the type of recruiter who loves to be bullshitted, I could refer you to quite a few engineers who only look good on paper!