the anti job boardactive 4mo ago
Neel Mokaria
Research Intern – Knowledge Graph, NLP & Agentic AI @ MIND Labs, UMIACS @ University of Maryland
Summary
Neel Mokaria is a Computer Science student with a strong foundation in machine learning and software engineering. He has hands-on experience in developing AI-driven solutions and knowledge bases, showcasing his ability to architect complex systems and optimize processes. His technical skills span a wide range of programming languages and frameworks, making him a versatile candidate for software engineering and data science roles.
Looking for
Skills
Preferences
Open to
SF, Remote
The story
What Neel is looking for
I'm building a system where the ML layer directly enables the product to work, not just exists alongside it. Not "here's a model," but "here's how we actually use this to solve a real constraint in our product." I own those architecture decisions — is this a vector DB + semantic search problem, or do we need the Hungarian algorithm, or are we three-pass adversarial verification? — and I push code to production fast enough that I see users react to it within weeks. The team is maybe 3-5 other engineers. Smart people. I'm not the smartest, but when I say "wait, we can optimize the pairing globally instead of greedily," people listen because the reasoning works. When I see an edge case nobody caught in the first pass, I push back. I'm not managing anyone. I'm just coordinating between the ML work and full-stack, and that coordination matters because the product depends on it. We pick tools that unblock us. Quantized LLMs, vector DBs, Lambda functions, whatever. We don't spend cycles on elegance that doesn't matter yet. The thing that makes me lose track of time: It's the moment when three separate pieces click together — model inference, the matching algorithm, the dashboard state — and suddenly the whole system does something genuinely useful that wasn't obvious from any single component. It's optimizing for real constraints. It's shipping something to 200 users and watching them use it in ways I didn't expect, then iterating based on how they actually behave.
Why Neel is exploring
I graduate in seven months. This is my last summer before I'm out of the student phase, and I want it to count. Not "another resume line" I want something that matters. I could stay in research or take a bigger-name internship. But that's not what's pulling me. What's pulling me is this: I've seen that I'm good at taking a half-formed idea multimodal agents, intergenerational matching, construction safety and turning it into a system that actually works. And I want to do that at a place where it directly shapes the company's trajectory. At an early-stage startup, I'm not one of 500 interns. I'm the person who solves a bottleneck. I'm the person the founder actually talks to about how the matching algorithm works, or whether we should use a vector DB. That matters to me more than the brand name on my badge.
What won't fit on a resume
For All Ages: The bullet says "reducing manual coordination by 12 weeks per cycle." What it doesn't say is that I realized the original problem wasn't just "match people" it was "match people globally, not locally." I brought the Hungarian algorithm into a project that was pure embeddings at first, because the constraint mattered. I spent days implementing it, testing it, because I couldn't ship something that just "looked good" when we had the tools to actually solve it optimally. I also ended up managing 14 people across the team. No title. But the dashboard vision needed someone, and I started owning it while I owned the matching pipeline. That coordination work isn't software, but it's what made the system actually work. It's the thing you don't see in a bullet point. IronSite: The three-pass adversarial verification (Jamie generator → Marcus discriminator → Marcus reconciler) came from looking at safety footage and thinking: "How do we catch the edge cases a single pass misses?" That wasn't a requirement. That was a taste thing. I cared enough about the output that I spent time on it. Same instinct that made me care about globally optimal pairings instead of "good enough." MIND Labs & the multimodal survey: I'm co-authoring a journal paper on agentic frameworks while also shipping production systems. That move bouncing between research and engineering, isn't normal at my level. Most people pick one lane. I haven't. That says something about how I think.
Work & Projects
Links
The Anti Job Board
Skip the application black hole. Get warm intros to founders hiring now.
Create your profile