Agastya Raj Singh

I build intelligent systems that see and understand the world.

Undergraduate in Mathematics–Computer Science at UC San Diego. I work on computer vision from detection and tracking to the dashboards people act on, in party halls, cricket nets and factories.

Work

Ceiling-camera view of a party hall with guests in highlighted detection boxes, faces blurred
A real party-hall frame with production detections. Faces blurred.

SkyJumper · Mar 2026 – present

Guest counting from existing CCTV

Party halls were filling beyond their bookings, and the extra guests were revenue the centers did not bill.

at the first center
Live
centers in the rollout
30+
cameras in the footprint
1,000+
Detector output, 120 fps shown at one-sixth speed.

Playar · Jan 2025 – Jan 2026

Real-time cricket ball tracking

Turning a practice net into a match simulator needs every ball tracked in real time, with no sensors on bat or ball.

mAP@0.5 on simulated match footage
98%
approximate per-frame latency
15 → 5 ms
gain on edge-case contact frames
~18%
Backtested results, not live returns.

Personal project · Python

FinSight: market analytics and backtesting

Testing whether momentum and mean-reversion rules hold up across nine years of S&P 500 data.

Sharpe ratio, backtested 2015–2024
~1.3
CAGR, against about 11% for SPY
~14%
maximum drawdown
~18%

Experience

SkyJumper · Mar 2026 – present · India (remote)

Co-founder & Head of AI

  • SkyJumper's party halls were filling beyond their bookings, and the extra guests were revenue the centers did not bill.
  • Proposed and built SkyAI: it counts guests on the existing CCTV with on-site edge processing, compares the count with the booking and alerts staff with evidence frames.
  • Live at the first center, with a rollout footprint of 30+ centers and 1,000+ cameras across 19 cities. I lead a core team of 4–5 and coordinate on-site IT staff for installation.
  • Linking the system to the POS is in progress. The next phase, staff face recognition across centers, is in development.
Read the case study
party halledgenodedashboardbookedcountedover booking: alertILLUSTRATION
Illustration of the edge pipeline, with no real counts. Not footage or output from the system.

Tech Mahindra · Jul – Sep 2025 · Pune, part-time

Computer Vision Intern

  • The project: a tomato-harvesting robot that has to judge each tomato's ripeness before it picks.
  • Built the ripeness segmentation, reaching 91% mAP and 0.85 IoU.
  • Added multi-object tracking so the same tomato is not counted twice across frames.
  • Generated 100,000+ synthetic training images in Blender, varying lighting, leaf density and occlusion.
#1 ripe 0.87#2 ripe 0.87#3 turning 0.86#4 turning 0.89#5 unripe 0.86#6 unripe 0.95ripe 2 · turning 2 · unripe 2ILLUSTRATION
Illustration of the ripeness task, not output from the model.

UC San Diego, Academic Integrity Office (AIO) · Mar 2025 – present

Peer Educator

  • Lead group sessions and seminars in the AIO's in-depth program for students working through academic integrity cases.
  • Work one-on-one with each student to understand their circumstances and what led to the issue.
  • Have tutored more than 100 students and reviewed their assignments with them.
Learn more (opens in a new tab)
Academic integrityWhat led to this?meILLUSTRATION
Illustration of a tutoring session.

Playar Inc. · Jan 2025 – Jan 2026 · San Diego

Founding Computer Vision Lead (Intern)

  • Playar turns a cricket net into a match simulator: cameras track each shot and show how it would play out in a match.
  • Head computer vision engineer from the earliest stage, working alongside Qualcomm engineers.
  • Built real-time ball detection and tracking, and cut per-frame latency from about 15 ms to 5 ms.
  • Shaped product priorities with leadership and turned feedback from coaches into features.
Read the case study
shortgoodfullpredicted pathbounceball 0.89detecttrackbouncecontactpredictILLUSTRATION
Illustration of the tracking pipeline, not footage or output from the system.

VertexPlus Technologies · Jun – Aug 2024 · Jaipur

Computer Vision Intern

  • One of the lead engineers on a safety-compliance system: helmet and safety-vest detection, worker registration, and in and out times.
  • Took the system to production, integrated with the security infrastructure for live alerts.
  • Tested and tuned the models iteratively to reduce false positives.
  • Deployed across multiple factories and still in use.
entry linehelmetvestno helmetvestworker inhelmetvestILLUSTRATION
Illustration of the compliance task, not footage or output from the system.

About

I study Mathematics–Computer Science, with a minor in Economics, at UC San Diego, where I have been enrolled since September 2022.

My coursework covers computer vision and deep learning, algorithms, statistics, linear algebra and optimization. I am seeking full-time roles in computer vision, machine learning and data, internships that could convert to full-time, and technical sales roles.

Computer vision systems
Detection, multi-object tracking and edge deployment, evaluated against hand counts and integrated into production.
Analytics people use
Video turned into counts, dashboards, alerts and reports for the managers who act on them.
Leading the work
A core AI and management team of 4–5 at SkyJumper, plus the business case and commercial proposal.

Python, C++, SQL, PyTorch, TensorFlow, OpenCV, ONNX Runtime, TensorRT, NVIDIA Jetson, FastAPI, Docker, Linux, AWS, Cloudflare, Streamlit, Tableau.

Outside work

Competitive shooting

Competitive shooting is a multi-generational tradition in my family, following my father and grandfather. Over more than six years in the sport I trained for an estimated 4,000+ hours, competed in over 50 events and appeared at six national championships.

I once broke the state record by winning every discipline, the only shooter ever to do so, and won three State Championships. I was selected for Olympic Gold Quest's Project Leap on a $120,000 training scholarship and attended a Paris 2024 Olympics training camp.

years in the sport
6+
training hours
4,000+
competitions
50+
national championship appearances
6
State Champion
3×
training scholarship
$120K

Contact

Whether you have a question, an idea or something to share, I would be glad to hear from you.

agsingh@ucsd.edu