Parthiv Dholaria

01

About Me

I'm an ML Engineer at Ripik.Ai with a B.Tech in Computer Science from IIIT Delhi (Class of 2025). I specialize in building production-grade computer vision systems — from model training and fine-tuning to deployment and monitoring at scale.

My work spans YOLOv8/SAM-based detection and segmentation, classical image processing, and model serving via TorchServe. I've deployed ML systems across AWS EC2 and on-premise Windows servers, maintaining robust process monitoring with >95% uptime.

I've had the opportunity to build solutions for major clients including Adani, JSW, JSPL, Sun Pharma, and international companies like CSN Brazil and Cemex Mexico.

                        1+
                        Year of Experience
                    

                        6+
                        Production Pipelines
                    

                        >95%
                        System Uptime
                    

                        8+
                        Enterprise Clients
                    

02

Work Experience

ML Engineer

Ripik.Ai

May 2025 — Present

Optimized Ripik Optimus to enhance production planning and manpower scheduling for Sun Pharma, reducing scheduling time by 20% using novel linear-time complexity algorithms.
Built an end-to-end Computer Vision sizing pipeline using YOLOv8 and SAM served via TorchServe to determine percentage composition of materials on conveyor belts for clients including Adani, HZL, JSPL, JSW, CSN Brazil, and Cemex Mexico. Achieved >90% alignment with lab data, improving blend composition accuracy.
Built a ferrous content prediction pipeline for BirlaWhite Cement using classical image processing (HSV masking, Otsu thresholding, CLAHE), achieving >90% accuracy against ground truth measurements.
Built a workforce safety monitoring pipeline using fine-tuned YOLOv8 (trained on Roboflow-curated datasets) to detect helmets, gloves, and PPE kits in high-risk hazardous zones. Achieved a true alert rate of >95% for safety violations.
Built a solar panel defect detection system using YOLOv8 segmentation with Ultralytics tracking on drone-recorded video for PGP Glass. Identified 90% of true defects (cracks, hotspots).
Led deployment hardening across AWS EC2 (Linux) and on-premise Windows servers. Implemented watchdog-based process monitoring with systemd (Linux) and NSSM (Windows) for automatic restart on failure, achieving >95% uptime across all deployed pipelines.

YOLOv8 SAM TorchServe OpenCV AWS EC2 Docker Roboflow systemd NSSM

ML Intern

NX Block Trades Pvt. Ltd.

Jan 2025 — May 2025

Analyzed high-frequency monthly options trading data and engineered systematic trading strategies with full backtest pipelines.
Designed a Quantile Regression Network to generate ON/OFF signals for strategies at the start of each trading day, resulting in a 10% increase in cumulative PnL.

Python PyTorch Quantitative Finance Backtesting

03

Research Experience

Research Intern

CISPA, Germany | TU Dortmund — Dr. Christian Rossow

May 2024 — Jul 2024

Investigated vulnerabilities in TLS handshakes and developed an ML-based detection system for server-side attacks, increasing detection rate by 18%.
Automated attack simulations across diverse server configurations and built an end-to-end pipeline for dataset curation, model training, and evaluation.

Network Security TLS Machine Learning Python

Undergraduate Researcher

MIDAS Labs, IIIT Delhi — Prof. Rajiv Ratan Shah

Jan 2024 — May 2024

Worked on ASVSpoof 2019 genuine vs. spoofed speech detection. Replicated one-class learning approaches combined with ResNet18 deep architectures for classification.

Speech Processing ResNet18 One-Class Learning PyTorch

04

Projects

CSIRO Biomass Estimation

Kaggle Competition · Jan 2025

Predicted pasture biomass components (dry total, green, clover, dead matter) from field images using a DINOv2-Large backbone with 2×2 tiled processing, spatial position embeddings, and multi-head attention aggregation. Applied physics-constrained outputs to enforce compositional consistency.

Ranked top 48% (1822/3804 teams)

DINOv2 CLAHE Attention Ensemble TTA

NumerAi: Crowd-Sourced Hedge Fund

Competition · Nov 2024

Predicted stock returns on live data with varying market features and concept drift using metric-based meta-learning (Matching Networks, Prototypical Networks).

Target correlation: 0.0221

Meta-Learning Matching Networks Prototypical Networks PyTorch

Evaluating Mathematical Reasoning of LLMs

Research Project · Nov 2024

Benchmarked Mixtral-8x7b, LLaMA-3.1-8b, and Gemma-9b on SVAMP, GSM8k, and GSM-Hard datasets across multiple prompting strategies (ZeroShot, CoT, ReAct, RAG). Developed a novel multi-agent framework combining specializer and master agents with RAG.

Up to 100% accuracy on SVAMP with CoT

LLMs RAG Multi-Agent CoT Prompting

ARC: Abstract and Reasoning Corpus

Research Project · Aug 2024

Developed two novel approaches for the ARC challenge: brute-force DFS over DSL function permutations (optimized via DP), and an LLM-guided heuristic search.

LLM-guided approach achieved 2x solve rate vs. brute-force

LLMs DFS Dynamic Programming DSL

05

Blog

Kaggle Competition Jan 2025

My Experience in the CSIRO Biomass Estimation Kaggle Competition

A deep dive into predicting pasture biomass from Australian grassland images — the challenges of small datasets, physical label constraints, missing metadata, and how we tackled them with teacher-student distillation and DINOv2.

DINOv2 Distillation CLAHE K-Fold

Read article

06

Technical Skills

Languages

Python C++ C Java

ML / Deep Learning

PyTorch Ultralytics YOLO SAM GDINO LangSAM OpenCV HuggingFace TorchServe Roboflow

Cloud & Infrastructure

AWS EC2 GCP Azure Docker Linux systemd NSSM

Tools & Frameworks

Git Django MongoDB MySQL Conda RTSP Streaming

Domains

Computer Vision Object Detection Segmentation Image Processing NLP GenAI (RAG) Agentic SDK

07

Education

B.Tech, Computer Science and Engineering

Indraprastha Institute of Information Technology, Delhi (IIIT-D)

2021 — 2025

8.42 CGPA

Dean's Award for Academic Excellence (3rd Year)

Member, Google Developer Student Club (GDSC), IIIT Delhi

08

Get in Touch

I'm always open to discussing new opportunities, interesting projects, or collaborations. Feel free to reach out!

Email parthiv.dholariya9221@gmail.com Phone +91-9313054893 GitHub parthivdholaria LinkedIn Parthiv Dholaria

ML Engineer @ Ripik.Ai