Abhi Desai
Hey, I'm Abhi. Currently a grad student @ USC. CS alum @ Penn State. Most recently, I built findwell. It's an app + iMessage agent for saving social media posts and easily finding them later. Check it out on the app store. Before that, agent benchmarking at GBOX, embedded ML at Infineon, and computer vision at Synaptics. Currently looking for applied AI roles where I can have high ownership from day 1.
Get in touch:
abhijaydesai@gmail.com
Links
Experience
-
2026 – Present
Founder at Findwell
Solo engineer. Hybrid search over saved posts, an iMessage agent, a React Native app.
-
2026
Software Engineer at GBOX.ai
Model-agnostic agent benchmarking framework and evaluation environments across mobile, web, and desktop.
-
2025
Embedded ML Engineer at Infineon
Radar gesture recognition and Wi-Fi presence sensing on PSoC MCUs. Real-time C on FreeRTOS.
-
2024
Computer Vision Intern at Synaptics
Teacher-student pipeline: zero-shot vision-language models distilled into detectors that run on Synaptics silicon.
-
2023 – 2024
Undergraduate Researcher, Computational Biology Lab at Penn State
Sequence alignment. Built a Levenshtein automaton that took a pairwise comparison from O(MN) to linear.
Education
-
2025 – 2027
MS Computer Science at USC
GPA 3.9. Deep Learning, Large Language Models, NLP, Computer Vision.
-
2021 – 2025
BS Computer Science at Penn State
GPA 3.8.
Extra
Projects
| Project | What it is |
|---|---|
| Findwell | Text it a post, ask for it later. BM25 + pgvector with rank fusion and reranking; MRR@3 from 0.69 to 0.99 on a hand-labeled set. |
| Agent Gary | Phone-based voice agent that completes real tasks for older adults. Origin Weekend × Replit finalist; led a four-person team. |
| Agentic video editor | CLI editor on LangGraph using Twelve Labs video understanding for semantic search over footage. |
| Predictive Coding | Backprop vs. predictive coding on the same conv encoder-decoder, tested as a Fashion-MNIST denoiser under noise and translation. JAX, Equinox, PCX. |
| AGI Fund | Pipeline that pulls stock data and technical indicators, runs economic events through an LLM, and builds analysis prompts from both. |
| GAN from scratch | PyTorch, no library shortcuts, to see min-max optimization fail and then work. |