Piyush S. — AI Researcher

I am a Data Scientist and AI Engineer with experience designing and deploying Large Language Model systems, ML pipelines, and intelligent agentic tools across NLP, computer vision, and analytics domains. I enjoy building end-to-end AI products that combine research-driven methods with real-world scalability. My work ranges from developing agronomist-facing LLM assistants for farmers funded by the Gates Foundation, to creating multilingual intent and entity recognition frameworks for enterprise chatbots, to designing transformer and graph-based models for bias detection in LLM outputs. I’ve also built optimisation algorithms for digital advertising, custom ML pipelines, data engineering systems, and educational AI tools that simplify complex concepts for developers. I am recently finished my MSc in Data Science at The University of Manchester, where I have built a multitask DeBERTa-based bias detection model and a full social listening platform integrating real-time social data with LLM-powered competitor analytics. Alongside my academic and industry work, I enjoy building ambitious personal projects. These include an AI Operating System (MIRA) that autonomously manages system operations via natural language, and TIE (Tool Integration Engine), a graph-based framework enabling LLMs to intelligently route and call tools using knowledge graph reasoning and GNNs. I also won the StarkWare Bounty at AI Encode London 2025 for an AI-driven blockchain game project. I am passionate about ethical AI, scalable architectures, and crafting systems that genuinely help people. My goal is to continue building advanced AI platforms, agentic systems, and LLM applications that push the boundary between automation and intelligence.

Stackforce AI infers this person is a Data Science and AI Engineering professional with a focus on scalable AI solutions.

Experience: 2 yrs 9 mos

Skills

Large Language Models (llm)
Generative Ai
Data Science
Machine Learning
Research And Development (r&d)
Embedded Systems
Mechanical Engineering

Career Highlights

Designed multilingual NLU framework improving accuracy by 25%
Built LLM-powered assistant reducing search time by over 60%
Developed bias detection model during MSc with real-time analytics

Work Experience

DeHaat

AI Consultant (3 mos)

Data Scientist (7 mos)

NuvoRetail

Junior Data Scientist (6 mos)

Indian Institute of Technology, Delhi

Research Assistant (4 mos)

Knowledge Solutions India

Machine Learning Intern (2 mos)

Indian Railways

In-plant Trainee (0 mo)

Larsen & Toubro Defence IC

Project Management Intern (1 mo)

NH Motorsports

Steering Committee Head (1 yr 4 mos)

Education

Master of Science - MS at The University of Manchester

A1 at Goethe Institut Indien

Bachelor of Engineering - BE at University of Mumbai

Piyush S.

AI Researcher

United Kingdom2 yrs 9 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Designed multilingual NLU framework improving accuracy by 25%
Built LLM-powered assistant reducing search time by over 60%
Developed bias detection model during MSc with real-time analytics

Stackforce AI infers this person is a Data Science and AI Engineering professional with a focus on scalable AI solutions.

Contact

Skills

Core Skills

Large Language Models (llm)Generative AiData ScienceMachine LearningResearch And Development (r&d)Embedded SystemsMechanical Engineering

Other Skills

Retrieval-Augmented Generation (RAG)AI AgentsFastAPIPython (Programming Language)Optical Character Recognition (OCR)Deep LearningRLHFReinforcement LearningFront-End DevelopmentBack-End DevelopmentDocumentationFlaskPostgreSQLReact.jsAlgorithms

About

Experience

2 yrs 9 mos

Total Experience

8 mos

Average Tenure

Current Experience

Dehaat

2 roles

AI Consultant

Feb 2025 – May 2025 · 3 mos · Gurugram, Haryana, India · Remote

Designed a multilingual NLU framework (intent + entity) integrated into DeHaat’s chatbot ecosystem, improving query understanding accuracy by ~25% compared to the previous rule-based version.
Built domain-specific datasets and annotation schemas in Hindi, English, and regional languages, enabling scalable onboarding of new agricultural intents across product lines.
Led architectural planning for future chatbot services, reducing model retraining time by 30–40% through modularized components and reusable embedding pipelines.
Delivered documentation and benchmarking guidelines used by internal ML teams to standardize evaluation and rollout of new models.

Large Language Models (LLM)Generative AIRetrieval-Augmented Generation (RAG)AI AgentsResearch and Development (R&D)FastAPI+3

Data Scientist

Jan 2024 – Aug 2024 · 7 mos · Gurugram, Haryana, India · On-site

Built an LLM-powered agronomist assistant used internally to provide advisory to farmers, reducing manual search time for crop disease & soil-management information by >60%.
Developed RLHF and DPO fine-tuning pipelines for open-source chat models, improving internal evaluation scores by ~20–30% and reducing model hallucination in domain-specific queries.
Created Flask APIs to productionize the assistant, enabling frictionless model deployment across frontend and backend teams with zero downtime rollouts.
Automated data extraction using Selenium + BeautifulSoup, generating 10,000+ agricultural records from government advisories and agronomy portals, forming the foundation of a VectorDB retrieval layer.
Integrated LangChain-based workflows and custom tools, speeding up experimentation cycles for prompt and agent iterations by ~35%.

Large Language Models (LLM)Generative AIData ScienceResearch and Development (R&D)RLHFReinforcement Learning+7

Nuvoretail

Junior Data Scientist

Jul 2023 – Jan 2024 · 6 mos · New Delhi, Delhi, India · On-site

Built and deployed transformer + random forest models to optimize search advertising, achieving 73% R² and enabling the marketing team to make 30–40% more accurate spend decisions.
Developed a dynamic budget allocation algorithm that reduced overspending on low-performing keywords by ~20%, improving ROI across Amazon ad campaigns.
Engineered a scraping pipeline using Selenium to collect 50,000+ Amazon product listings, which became the primary dataset for market share analysis and ML experimentation.
Produced actionable insights on SKU-level performance, empowering leadership to adjust pricing and bidding strategies based on market competitiveness.

AlgorithmsData AnalysisPandasMicrosoft OfficeData StructuresMicrosoft Excel+19

Indian institute of technology, delhi

Research Assistant

Apr 2022 – Aug 2022 · 4 mos · Delhi, India · On-site

Led an ICMR-funded research project to build a nerve monitoring prototype using custom microcontroller firmware, reducing latency of signal acquisition by ~35%.
Designed algorithms to assess skin impedance data, improving system stability and accuracy across varying contact conditions.
Delivered technical reports, calibration results, and demo presentations, forming the basis for follow-up grant submissions and lab-scale deployments.

AlgorithmsResearch and Development (R&D)Industrial electronicsEmbedded SystemsPython (Programming Language)C+++4

Knowledge solutions india

Machine Learning Intern

Sep 2021 – Nov 2021 · 2 mos · Remote · Remote

Developed SQL pipelines to extract and transform large relational datasets, reducing manual reporting time by >50%.
Built regression and clustering models for client datasets, improving feature interpretability and decision-making for internal stakeholders.
Conducted EDA and feature engineering that increased baseline model performance by ~10–15%.

Data AnalysisPython (Programming Language)Data ScienceK-Nearest Neighbors (KNN)Machine LearningLinear Regression+3

Indian railways

In-plant Trainee

Dec 2019 – Dec 2019 · 0 mo · Mumbai, Maharashtra, India

Learned the working and maintenance of trains and it's management system

Documentation

Larsen & toubro defence ic

Project Management Intern

Jun 2019 – Jul 2019 · 1 mo · Mumbai, Maharashtra, India

Updated and maintained Bill of Materials using Excel automation, improving engineering team visibility on inventory and reducing mismatch errors by ~30%.
Researched hydraulic systems and produced a comprehensive technical report, supporting training material for junior engineers.

Microsoft OfficeMicrosoft ExcelMicrosoft WordDocumentation

Nh motorsports

Steering Committee Head

Jan 2019 – May 2020 · 1 yr 4 mos · Thane, Maharashtra, India

1. Co-Founder of Society of Automotive Engineers (SAE) club at New Horizon Institute of Technology and Management.
2. Led the steering department
2. Planned, designed and fabricated steering system for the Formula Student Car at NH Motorsports
3. Supervised Sponsorship department and guided members for sponsorships.
4. Brought sponsorships from various sources

SOLIDWORKSMechanical Engineering