Open to conversations New Delhi, India Liat AI · Remote

Lakshay Bhatia Deep Learning Engineer

I build reliable machine learning systems through disciplined experimentation, clear engineering, and measurable outcomes — computer vision for sports analytics, agentic LLM platforms, and the pipelines that keep both alive in production.

Role
Deep Learning Engineer
Current
Liat AI · New York, remote
Focus
Computer vision · Agentic AI · MLOps
Approach
Research to production

§ 01 / About

Profile

Illustrated portrait of Lakshay Bhatia
Lakshay Bhatia · New Delhi

I’m a deep learning engineer who enjoys turning ambiguous problems into systems that are useful, testable, and dependable.

My work moves between experimentation, data, models, and the engineering required to make them hold up in practice. I focus on whether the result survives contact with real data.

Most of what I build lives in computer vision — detection, tracking, and keypoint models for sports analytics — with side work in agentic LLM systems and the MLOps plumbing that keeps either one running.

That means understanding failure modes, running disciplined experiments, keeping work reproducible, and communicating trade-offs clearly.

Based in
New Delhi, India
Working with
Liat AI · Sports analytics
Previously
HawkVision AI · Production CV systems
Education
DCRUST · IIT Madras

§ 02 / Experience

2 positions

Aug 2025 — Present

New York, USA · Remote

Liat AI

  • Deep Learning Engineer

    Apr 2026 — Present

  • Deep Learning Intern

    Aug 2025 — Mar 2026

  • Built and shipped vision systems across jersey, card, keypoint, and counter-attack detection, along with a translation pipeline.
  • Resolved a critical detection failure, improving F1 from 0.45 to 0.81 through structured experiments across models, data, and interpolation strategies.
  • Built evaluation and training workflows on 50,000+ hand-annotated frames, plus 1,800 video clips labelled in CVAT for a spatio-temporal actor detector.
  • Led the initial research into highlight generation, porting the systems to another sport, DevOps and production pipeline design, and market and competitor analysis.
15+Models shipped
50K+Frames annotated
5Systems built
+80%F1 lift

Jun 2025 — Jul 2025

Jaipur, India

HawkVision AI

Deep Learning Intern

  • Shipped three computer-vision systems in six weeks across safety, traffic, and compliance workflows.
  • Built a CPU-first OCR pipeline reaching 95% character accuracy while reducing compute cost by 60%.
  • Improved per-stream video throughput from 8 to 25 FPS with thread pooling and queue-based buffering.
3.1×Throughput
95%OCR accuracy
−60%Compute cost
6Weeks

§ 04 / Toolkit

7 categories

Languages

  • Python
  • C++
  • SQL
  • Rust
  • Java

Machine learning

  • PyTorch
  • Scikit-learn
  • YOLO
  • RF-DETR
  • OpenCV
  • PaddleOCR

Concepts

  • Computer vision
  • MLOps
  • RAG
  • Agentic AI
  • LLM fine-tuning
  • MCP

Backend

  • FastAPI
  • Flask
  • Multithreading
  • Low-level design

Infrastructure

  • Docker
  • Git
  • Linux
  • AWS (EC2, S3)
  • MLflow
  • ZenML
  • Opik

Data

  • Pandas
  • NumPy
  • Matplotlib
  • Seaborn

Databases

  • PostgreSQL
  • MySQL
  • MongoDB

§ 06 / Credentials

Education · Certifications · Leadership

Education

Nov 2022 — Jun 2026

Deenbandhu Chhotu Ram University of Science and Technology

B.Tech · Computer Science · CGPA 7.9 / 10

Jan 2023 — Jun 2024

Indian Institute of Technology Madras

B.S. coursework · Data Science · CGPA 8.0 / 10

Certifications

2024

Deep Learning Specialization

DeepLearning.AI

2024

Machine Learning Specialization

DeepLearning.AI

In progress · 5 of 8

Google Cybersecurity Specialization

Google

Leadership

Google Developer Groups

Campus Lead

Led a community of 300+ students across 12 bootcamps, growing active developers by 50%.

GDSC · Robotics Society

Core member

Ran 10+ workshops on machine learning and robotics; mentored 15 students through capstones.

CodeChef · Codeforces

Competitive programmer

3★ on CodeChef · 1189 rating on Codeforces.

§ 07 / Contact

New Delhi · UTC+5:30

Have a hard problem?

Tell me what you are trying to build, where it is failing, and what a good outcome looks like.

lakshay@lakshaybhatia.com