ManasVellaturi

CS + STATISTICS, UNC CHAPEL HILL

AI/ML Engineer, Software Engineer, Computer Vision Researcher, Full-Stack Developer, Passionate Builder

EducationUNC Chapel Hill2025 – 2029

FocusEvent-camera research at UNC

Open toI'm open to research collaborations, internships, and interesting problems.

I'm an AI/ML developer building systems from event-camera hardware to a live product — computer vision, NLP, and the connectome as neural architecture.

View Projects Download Resume

SHIPPED AND LIVE

Janora — AI Agent Readiness Platform

Teams are putting AI agents into real workflows faster than they can answer whether a given one is safe to turn on. Janora is my v1 attempt at making that a decision with evidence behind it rather than a judgement call — it tests an agent against a specific workflow and produces a Launch Readiness Report that says go or no-go.

AI/MLWeb

janora.dev

Case study

WHAT I AM WORKING ON

Now

Event-camera research at UNC

Building analog hardware and Python tooling for dynamic vision sensors — a field I had no background in when I started.

C. elegans connectome as an RL policy

Training an agent whose policy network is a real, fully mapped nervous system. In development toward a conference submission.

Janora v1

Live at janora.dev, and still being built out.

A LITTLE ABOUT THE WORK

About

I'm double-majoring in Computer Science and Statistics & Analytics at UNC Chapel Hill, and I build AI/ML systems end-to-end — from the hardware that produces the data through to the product someone actually uses.

In practice that has meant fairly different things in the same year. Wiring op-amp and MOSFET circuits for an event camera in a research lab. Implementing the only completely mapped nervous system we have as the policy network of a reinforcement learning agent. Shipping a working product to a live domain on my own. The common thread is that I'd rather build the whole path than one layer of it.

Most of that was work I hadn't done before. I started the event-camera research with no background in analog electronics and learned the circuit design and validation methodology on the way through; Janora was the first product I'd taken from an idea to something running in public. When I want to be sure I actually understand something I build it by hand and then write it up — my NLP pipeline exists because I wanted to know how models really work, and the two beginner guides I published came out of it.

TOOLS & TERRITORY

Skills

Languages

PythonJavaSQLJavaScript

AI & ML

PyTorchscikit-learnReinforcement Learning (PPO)NLPComputer VisionPrompt EngineeringGoogle Gemini

Web

React.jsNode.jsFull-Stack DevelopmentAPI Development

Research & Systems

Event-Based SensingNeuromorphic VisionRaspberry Pi GPIOAnalog Circuit DesignData PipelinesSoftware Testing

WHERE THE HOURS WENT

Experience

01 · Nov 2025 – Present

Undergraduate Student Researcher

UNC Department of Computer Science · Chapel Hill, NC · On-site

I joined this lab with no background in analog electronics. I now conduct event-camera research using dynamic vision sensors (DVS/EVS) to capture asynchronous visual data for high-speed computer vision experiments. I built and wired precision analog hardware — a constant-current infrared LED driver circuit (op-amp/MOSFET-based), photodiode-based optical feedback for closed-loop validation, Raspberry Pi GPIO control, PT4115 constant-current LED drivers, and synchronized camera trigger signals. I developed Python tools for programmable LED pulse generation with adjustable pulse width, frequency, and duty cycle, plus data-processing pipelines for trigger-window extraction, pulse-level event analysis, and quality filtering of noisy recordings. I built diagnostic visualizations — cumulative event curves, event-rate plots, and reconstructed event images — to validate recording quality and system behavior, learning analog circuit design and validation methodology in the process. I collaborate with a research team on neuromorphic vision, event-based sensing, and high-temporal-resolution visual data analysis.

02 · Dec 2025 – Present

Vice President of Company Outreach

UNC IEEE · Raleigh-Durham-Chapel Hill Area (Member since Nov 2025)

I lead corporate outreach and sponsorship strategy, building and maintaining relationships with industry partners to strengthen IEEE's professional network. I organize company info sessions and technical talks for UNC's engineering community.

03 · Aug 2022 – Jun 2025

Vice President → Co-Vice President

Initiative Programs · Panther Creek High School

I led monthly meetings for 30+ high school tutors, enhancing their tutoring skills through constructive feedback and assistance, while managing tutoring initiatives at Panther Creek High School.

RECENT BUILDS

Projects

Cadent — AI Study Planner for College

Jul 2026 – Present

· Live · In progress

I'm the creator and sole developer. I built a macOS desktop app that reads a student's real Canvas coursework — assignments, syllabi, announcements, and PDFs — and turns it into a daily study plan, entirely on-device. A grounded extraction layer, a capacity-aware scheduler, a conversational replanning layer, and a verification guard that reviews every change before it's committed all agree on one model of what an assignment is.

AI/MLDesktopEdTech

Private beta (invite-only)

Janora — AI Agent Readiness Platform

Jun 2026 – Present

· Live · In progress

I'm the creator and sole developer. I built a working v1 platform that tests whether AI agents are safe to launch for a given workflow, evaluating company truth, agent behavior, and action safety through workflow packs, agent manifests, generated test scenarios, and manual/transcript/sandbox testing. I also built the certification and Launch Readiness Report system that turns testing into a clear go/no-go decision.

AI/MLWeb

janora.dev

C. elegans Connectome as a Learnable RL Policy

Mar 2026 – Present

· In progress

Independent research: I use the complete C. elegans connectome (the only fully mapped nervous system) as the policy network of a reinforcement learning agent trained to perform chemotaxis. I built the simulation environment, implemented the connectome as a neural network from real connectivity data, and enforced biological constraints during training. Currently in development toward a conference submission with faculty feedback.

ResearchAI/ML

TextScope — NLP Pipeline & Model Optimization

Feb – May 2026

· Complete

I wanted to understand how NLP models actually work, so I built the pipeline by hand — tokenization, bigram extraction, TF-IDF feature engineering, and logistic regression tuned across L1/L2 regularization and cross-validation — then wrote up what I learned. Switching from TF-IDF to sentence embeddings raised cross-validated accuracy from 0.50 to 0.96.

AI/ML

InvestIQ — AI-Powered Investment Analysis

Oct 2025

· Complete

Built at HackNC. I integrated the Google Gemini API to generate AI-driven investment insights, engineered structured prompts to control response format, tone, and analytical depth, and built the frontend that surfaces those insights.

HackathonAI/ML

exoMatch — Exoplanet Data Exploration

Sep 2025

Built for Carolina Data Challenge. I built a web platform for exploring and comparing exoplanet datasets through interactive, visually structured UI components, focused on presenting scientific data clearly.

HackathonWeb

Club Management Application

Jun – Aug 2022

A full-stack club management app I built, frontend-focused with backend contributions supporting application functionality and data handling.

Web

WHERE IT STARTED

Education

2025 – 2029

UNC Chapel Hill

Chapel Hill, NC

B.S. Computer Science and B.S. Statistics and Analytics (double major) · Currently enrolled

Aug 2021 – Jun 2025

Panther Creek High School

Cary, NC

1520 SAT · 4.7115 W-GPA · Rank 7/606

SAY HELLO

Contact

I'm open to research collaborations, internships, and interesting problems.