Pranay Saha memoji

Hey, I'm Pranay 👋

Computer Vision & ML Engineer

Building scalable AI inference systems & city-scale surveillance platforms.
Currently at Masterworks, Hyderabad.

Scroll
01 — About me

Turning pixels
into intelligence

I'm a Computer Vision & Machine Learning Engineer with a deep focus on building production-grade AI systems that operate at scale. From managing 4,000+ camera feeds to reducing computational overhead, I engineer solutions that are both technically rigorous and practically impactful.

🎓
Education
B.Tech CSE (AIML) — IEM Kolkata
📧

Tech Stack

Python Java PyTorch TensorFlow TensorRT ONNX Triton Inference Server OpenCV LangGraph Docker Kubernetes Kafka FastAPI Git
02 — Experience

Where I've
made an impact

Masterworks
Computer Vision Engineer · SDE-1
Mar 2025 – Present Hyderabad, India
  • Contributed to the flagship low-code platform powering city-scale surveillance across highways, parks, and streets with automated incident detection and real-time alerting.
  • Architected multiple microservices enabling seamless one-click deployment of CV use cases across distributed edge devices.
  • Managed high-throughput model inference using Triton Inference Server, supporting 24×7 concurrent camera streaming across large deployments.
  • Built a dynamic multi-accelerator AI inference orchestration service converting models into optimised engines for NVIDIA and Qualcomm hardware.
  • Designed inference architecture processing real-time video feeds from 4,000+ cameras across heterogeneous edge devices.
  • Optimised end-to-end inference pipeline, achieving ~70% reduction in computational overhead.
  • Built user-configurable CV Use Nodes (Detection, Annotation, Classification, Alerting) enabling rapid workflow customisation.
  • Engineered data pipelines and trained CV models for city surveillance tasks including entry/exit monitoring and access alerting.
TritonTensorRTONNXDockerKubernetesKafka
Yukin
Machine Learning Engineer
Feb 2024 – Dec 2024 Remote
  • Collaborated with the founding team developing models for personalised Digital Avatars with realistic expressions and lip-synchronisation.
  • Evaluated 20+ open-source models for Text2Speech, Voice Cloning, and Speech Synthesis in a multilingual setup; optimised to reduce RAM consumption by 50%.
  • Designed and developed a RESTful API using FastAPI, improving developer productivity by 20%.
FastAPITTSVoice CloningPython
Proglint Solutions
Computer Vision Engineer
Contract / POC
  • Constructed a Proof of Concept (POC) for real-time person re-identification and multi-camera tracking, designed to enhance surveillance coverage in retail environments.
  • Engineered a highly precise model boasting a 98% precision rate, significantly slashing False Positives by 30%.
  • Streamlined algorithmic performance for enhanced speed and efficiency, reducing frame computation by 50%, leveraging advanced techniques such as multi-threading for seamless parallel processing.
Re-IDMulti-Camera TrackingOpenCVMulti-threadingPython
03 — Blog

Things I write
about

Inference · MLOps 5 min read

Why Companies End Up Using Triton Inference Server — A Simple Case Study

A deep-dive into why Triton Inference Server has become the go-to choice for production ML inference — exploring real bottlenecks, benchmarks, and the engineering trade-offs companies face when scaling model serving.

Pranay Saha →
Apple Silicon · ML 6 min read

Dive into MLX — Performance & Flexibility for Apple Silicon

A hands-on exploration of Apple's MLX framework — how it unlocks near-native GPU performance on M-series chips, its unique unified-memory model, and where it fits in the ML ecosystem.

Pranay Saha →
Read all on Medium →
04 — Contact

Let's build
something together

Whether it's an exciting AI opportunity, a collaboration, or just a chat about inference systems — my inbox is always open.