Safety
Tay
Microsoft Tay
Analyzing how an open AI chatbot launch failed when abuse prevention was not treated as core product behavior.
Product Sense
01
Case Study
GitHub Copilot Case Study
Why Developers Still Don’t Trust AI Code
AI coding assistant teardown focused on developer trust, onboarding, code acceptance, workflow friction, and confidence-building loops.
Read Full Case Study
Supabase Case Study
Helping Non-SQL Builders Create Production-Ready Backends
Backend setup teardown focused on schema creation, relationships, permissions, RLS confidence, and guided production readiness.
Read Full Case Study
Ollama Case Study
Choosing the Right Local AI Model Without Guesswork
Local AI workflow teardown focused on model choice, hardware fit, privacy tradeoffs, setup confidence, and guided recommendations.
Read Full Case Study
Razorpay Case Study
Making Settlement Decisions Clear for Merchants
Merchant finance teardown focused on cash-flow clarity, available balance, payout timing, expense visibility, and action-oriented decisions.
Read Full Case Study
02
Failure Case Study
Platform
Stadia
Google Stadia
Diagnosing why strong cloud gaming technology failed to become a trusted gaming ecosystem.
Utility
Humane
Humane AI Pin
Mapping where AI hardware overreached before proving one repeatable daily-use case.
Agents
Rabbit
Rabbit R1
Studying the gap between agentic product promise and launch reliability in everyday workflows.
03
Analysis Case Study
AI Support
Launch Ready?
Should an AI Support Agent Go Live?
Interactive analytical case study using readiness metrics, answer quality, model tradeoffs, and guardrail risk.
Benchmark
RAGLab
RAG Strategy Comparison
Evaluating retrieval strategy tradeoffs through side-by-side product experience design.
Agents
AdPilot
Agentic Campaign Simulation
Breaking down how multi-agent workflows can support marketing decisions before spend.
DevOps
DeployPilot
Deployment Confidence Loop
Analyzing how AI can make release readiness, infra risk, and next action more visible.
Prototyping Skills
DataTalk
Natural-language querying over company data. It uses a constrained intent/slot compiler for reliable supported queries, plus an experimental SLM path that learns schema and question patterns.
RAGLab
Retrieval benchmark lab comparing BM25, vector, hybrid, memory, and graph RAG on the same prepared datasets with visible evidence, latency, and method ranking.
AdPilot
Multi-agent campaign validator for online and offline marketing. It captures email, analyzes persona, geography, channel mix, creative risk, and forecasts launch metrics before spend.
DeployPilot
AI DevOps simulation lab for Harness-style CI/CD flows. It runs deployment scenarios, captures tester email, shows logs, explains failures, and saves run history.
Business Builder
OnlyScholar
Research agency built around AI, ML, Deep Learning, Computer Vision, GenAI, and Agentic AI papers.
25+
papers completed
Projected Growth
Built from a 2/month completed run rate toward a 10/month projected research delivery run rate.
21+ Journal Acceptances
Research projects moved through submission, revision, and acceptance workflows.
15+ Active Projects
Active research ideas across AI, computer vision, NLP, GenAI, and agentic systems.
SCI Paper Every Year
OnlyScholar simplifies topic selection, implementation, writing, submission, revision, and publication workflows.
AI Research Publication
Research Publications
Published work across applied AI, computer vision, human activity recognition, and intelligent safety systems.
2024
DATS
DATS: Driver Assistance Text Spotter
AIP Conference Proceedings · AIP Publishing LLC · 2024End-to-end scene text spotting system for driver assistance. The work focuses on detecting and recognizing blurry, sheared, noisy, and small text that appears in driving environments.
- Uses DenseNet-121 to generate higher-dimensional feature maps for scene text spotting.
- Introduces NTDnet, a Noisy Text Detection Network, to handle difficult text cases.
- Improves scene text spotting precision by 5.1% and shows at least 3.2% improvement across small, medium, and large text.
2023
LightHARS
LightHARS: A Light Human Activity Recognition System
Springer Nature Singapore · ICMISC 2022 Proceedings · 2023Privacy-aware human activity recognition system using sensor data instead of video-heavy inputs. The model is designed to reduce parameter count and computation time while keeping recognition performance strong.
- Builds a cost-efficient deep neural network for human activity recognition.
- Uses sensor data to reduce identity exposure compared with video-based HAR systems.
- Outperforms existing HAR models by achieving 4.5% higher accuracy.
2023
Insect Recognition
Insect Recognition: A Visionary Step Towards Smart Agriculture
AIP Conference Proceedings · AIP Publishing LLC · 2023Computer vision system for smart agriculture that identifies crop-threatening insects so farmers can take preventive action before attacks reduce production.
- Targets insect detection as a practical problem for drones, robots, and smart agriculture workflows.
- Achieves 81% recognition performance with a computationally effective model.
- Uses 1,640,423 parameters, keeping the system adaptable for existing agriculture equipment.
2020
Gas Leakage Security
Automated Gas Leakage Security System
Journal of Emerging Technologies and Innovative Research · 2020IoT-based toxic gas detection and alerting system for homes, offices, education spaces, and industrial environments. The system monitors gas levels continuously and sends fast alerts during leakage events.
- Designs a microcontroller-based gas detection and safety alerting system.
- Triggers alarm, email, SMS, and call alerts when toxic gas exceeds the normal level.
- Focuses on faster emergency response and reducing further leakage risk.
About Darsh
01
Experience
Job Experience
Jan 2026 - Current
SDE-II, Data Engineer
Refyne India: infra-level data systems, pipeline reliability, orchestration, monitoring, and platform quality.
Jan 2025 - Jan 2026
SDE-I, Data Engineer
Refyne India: scalable ETL pipelines using Python, Airflow, and Snowflake; improved query performance and monitoring.
Mar 2024 - Jan 2025
Associate SDE, Data Engineer
Refyne India: AWS ETL/ELT pipelines, Redshift workflows, reporting systems, and AI infrastructure support.
Sept 2023 - Mar 2024
Data Engineer Intern
Refyne India: Airflow workflows, SQL optimization, and in-house ETL framework components.
Dec 2021 - Nov 2022
Lecturer
Parul University: taught data structures, database systems, Python programming, deep learning labs, and guided capstones.
Business & Freelancing
July 2025 - Current
OnlyScholar
Research writing, AI implementation, topic framing, experiments, paper writing, submission, and publication support.
Jan 2023 - July 2025
Freelance Academic Research Paper Writer
Computer-vision implementation, literature surveys, methodology writing, results drafting, and publication-ready papers.
02
Education
Ph.D.
Information Technology
Indian Institute of Information Technology, Allahabad
- Working on architecture-driven AI research.
- Exploring scene text spotting through diffusion and architectural changes.
- Studying deep models from research and implementation angles.
Master of Technology
Computer Engineering
Parul University · Gold Medalist · CGPA 9.09 / 10.0
- Specialized in AI, machine learning, and computer vision.
- Received Gold Medalist recognition for academic performance.
- Published 3 research papers during the program.
Bachelor of Technology
Computer Science Engineering
Parul University · CGPA 7.19 / 10.0
- Developed a capstone project around IoT systems.
- Built early research exposure in deep learning.
- Organized 5 hackathons and startup-focused events.
03
My Mentors
Dr. Randheer Bagi
LinkedInHe mentors me to polish my research depth, AI fundamentals, and academic consistency. His guidance helps me sharpen research questions, model understanding, and publication discipline. He keeps my work anchored in rigor, contribution, and long-term learning.
Sachin Sharma
LinkedInHe helps me grow business judgment, product thinking, and scale-focused execution. His coaching connects customer problems, positioning, GTM, and growth mindset. He pushes ideas toward clearer product decisions, stronger business models, and measurable traction.
06
Let's connect