Prakshal Mehra

AI Researcher

Jaipur, Rajasthan, India6 yrs 8 mos experience

Most Likely To SwitchAI Enabled

Key Highlights

Achieved 97% accuracy in demand forecasting.
Designed end-to-end architecture for generative AI.
Reduced fraud costs by $50k YoY.

Stackforce AI infers this person is a Data Scientist specializing in AI and machine learning for SaaS and Fintech industries.

Contact

Skills

Core Skills

Machine LearningGenerative AiPredictive ModelingArtificial Intelligence (ai)Deep LearningComputer VisionConversational Ai

Other Skills

AIAWSAmazon EKSAmazon Web Services (AWS)Android StudioBERT (Language Model)BashC (Programming Language)C++Data Classification and ModellingData ScienceDjangoDockerFirebaseHTML

About

Results‐driven Data Scientist with expertise in machine learning, generative AI, and predictive modeling with proven success in demand and price forecasting along with RAG based applications, delivering high‐accuracy solutions and scalable tools. Skilled in the deployment of cloud‐based AI systems, real‐time monitoring, and automation pipelines. Keen to enhance my skills and stay updated with the latest trends.

Experience

6 yrs 8 mos

Total Experience

1 yr 2 mos

Average Tenure

2 yrs 8 mos

Current Experience

Valiance solutions

Senior AI Engineer

Oct 2023 – Present · 2 yrs 8 mos

Generative AI
Deployed several open source models and optimised it for max scalability with minimum latency.
Designed and deployed the end‑to‑end architecture for RAG‑based conversational and generative AI with multi‑lingual support using Multi‑Agents, LangChain and LlamaIndex.
Established an efficient pipeline for pre‑processing PDFs and multi‑modal, thereby generating an accurate data lake.
Developed and implemented real‑time LLM evaluation and monitoring service using Langfuse, UpTrain and DeepEval.
Demand & Price Forecasting:
Optimized Production Planning: Designed and developed machine learning and probabilistic modeling initiatives for B2B demand
forecasting of viscose fiber, driving $14.7M in business impact.
Achieved accuracy of 97% for high‑volume SKUs and 85% accuracy for low‑volume ones (intermittent demand)
Architected price forecasting using ML, DL and transformer based models for wool fiber and its global indicators and achieved 87% directional accuracy and 90% forecast accuracy.
Built an end‑to‑end automated tool for business users, providing 3‑month demand projections and identifying key drivers.
Enhanced project scalability with Dockerized modular codebases and introduced parallelized AWS batch processing for optimised resource utilisation.

Generative AIMachine LearningPredictive ModelingData ScienceAWSDocker

Fogteams

AI/ML Engineer (Founding Member)

Sep 2022 – Sep 2023 · 1 yr

Led AI/Data Science unit, undertaking multi-faceted responsibilities:
AI/ML engineer, product research, model deployment and management.
Developed Slack and Meet Co-Pilots to assist in retrieving insights from
meetings, managing workflow and tasks.
Generated transcript: real-time speech to text using whisper AI.
Retrieved Meet insights using Deep Learning, transformer models and
LLMs.
Designed Intent Classifier and Entity Extraction using DIET, BERT, and
Duckling model.
Processed the classified queries using SQL, GPT, and LangChain.
Deployed AI models on AWS and Digital Ocean using open source
streaming framework built on Apache Kafka (Airy).
Reduced string search time from 3 mins to 10-100ms in PgSQL, reduced
end-to-end latency by over 10%.

Artificial Intelligence (AI)Open-Source SoftwareLarge Language Models (LLM)BERT (Language Model)Deep LearningAWS+1

Citymall

Data Scientist 1

Sep 2021 – Aug 2022 · 11 mos

Implemented Fraud User Identification system : reduced false rejection rate drastically with 90% accuracy. Reduced fraud cost by $50k YoY.
Implemented Online-KYC using VIT to accurately classify documents. YOLOv5 for Object detection and character recognition via Transformer OCR was utilised.
Designed and deployed a data migration service, resulting in 25% speedup in scheduling pipelines for continuous data archiving and major load reduction.
Optimized DevOps team's workflow efficiency by integrating a Slack bot that automated the user access granting process for various services.

Artificial Intelligence (AI)Transformer ModelsComputer VisionDeep LearningSQLNLP