Prajwal Rudrapwar

About

Innovative AI/ML Engineer with 2+ years of experience architecting high-performance, real-time AI solutions and multi-agent pipelines. Proven track record of translating complex product requirements into low-latency production applications, including StreamGuard (WebSocket-driven live stream co-pilots), RAG-based legal document analysis engines, and aspect-based NLP engines. Expert in optimizing LLM inference (Gemini, Transformers), vector databases (pgvector), and real-time streaming architectures (WebSockets, voice matching). Passionate about building intelligent, scalable systems that solve real-world latency and user engagement bottlenecks.

Skills & Expertise (32)

Generative AI Advanced

9.0/10

2

Years Exp

Large Language Models Advanced

9.0/10

2

Years Exp

Hugging Face Transformers Advanced

9.0/10

2

Years Exp

LangChain Advanced

8.5/10

2

Years Exp

Vector Embeddings Advanced

8.5/10

2

Years Exp

Python Advanced

8.3/10

2

Years Exp

FastAPI Advanced

8.1/10

2

Years Exp

SpaCy Advanced

8.0/10

2

Years Exp

Pydantic Advanced

8.0/10

2

Years Exp

TensorFlow Advanced

8.0/10

2

Years Exp

scikit-learn Advanced

8.0/10

2

Years Exp

RESTful APIs Advanced

7.8/10

2

Years Exp

JWT-based Authentication Intermediate

7.5/10

2

Years Exp

Docker Intermediate

7.5/10

2

Years Exp

Voice Recognition Intermediate

7.5/10

2

Years Exp

Postgresql Intermediate

7.5/10

2

Years Exp

SQL Intermediate

7.5/10

2

Years Exp

Git Intermediate

7.5/10

2

Years Exp

Next.js Intermediate

6.5/10

2

Years Exp

React Intermediate

6.5/10

2

Years Exp

JavaScript Intermediate

6.5/10

2

Years Exp

ReportLab Web Scraping BeautifulSoup Httpx Tokenization RoBERTa Retrieval-Augmented Generation Supabase Row-Level Security Database Schema Design PGvector

Work Experience

AI/ML Engineer

Addvic Technology

Jul 2024 - Present

Architected and deployed an asynchronous AI-powered contract analysis engine using FastAPI and Supabase (PostgreSQL), reducing manual contract review times by 85%. Engineered a production-grade RAG (Retrieval-Augmented Generation) pipeline utilizing Gemini text-embedding-004 and Supabase pgvector, allowing users to perform semantic search and query long-form legal documents with sub-second response times. Implemented an advanced smart chunking algorithm that dynamically splits PDFs and DOCX files by legal sections (using PyMuPDF and LangChain), improving retrieval context accuracy by 40% compared to standard character splitters. Integrated Gemini LLM API to automate complex risk assessments, utilizing strict structured JSON output to extract party agreements, critical dates, and risk ratings (low/critical) with 98% data extraction integrity. Developed a professional PDF reporting service from scratch using ReportLab, enabling immediate downloads of structured executive summaries, risk matrices, and actionable clause recommendations. Engineered robust data pipelines for complex classification datasets, applying advanced feature engineering to improve baseline Machine Learning model accuracy by 25%. Implemented Model Pruning and quantization strategies on deep neural networks, reducing inference latency and memory footprint by over 40% without compromising predictive performance. Evaluated multiple classification algorithms and automated hyperparameter tuning using Scikit-Learn and TensorFlow, establishing highly accurate models for production readiness.

Machine Learning Intern

Kiran Academy

Jan 2024 - Jun 2024

Architected and deployed an asynchronous AI-powered contract analysis engine using FastAPI and Supabase (PostgreSQL), reducing manual contract review times by 85%. Engineered a production-grade RAG (Retrieval-Augmented Generation) pipeline utilizing Gemini text-embedding-004 and Supabase pgvector, allowing users to perform semantic search and query long-form legal documents with sub-second response times. Implemented an advanced smart chunking algorithm that dynamically splits PDFs and DOCX files by legal sections (using PyMuPDF and LangChain), improving retrieval context accuracy by 40% compared to standard character splitters. Integrated Gemini LLM API to automate complex risk assessments, utilizing strict structured JSON output to extract party agreements, critical dates, and risk ratings (low/critical) with 98% data extraction integrity. Developed a professional PDF reporting service from scratch using ReportLab, enabling immediate downloads of structured executive summaries, risk matrices, and actionable clause recommendations. Engineered robust data pipelines for complex classification datasets, applying advanced feature engineering to improve baseline Machine Learning model accuracy by 25%. Implemented Model Pruning and quantization strategies on deep neural networks, reducing inference latency and memory footprint by over 40% without compromising predictive performance. Evaluated multiple classification algorithms and automated hyperparameter tuning using Scikit-Learn and TensorFlow, establishing highly accurate models for production readiness.

Machine Learning Intern

YBI Foundation

Nov 2023 - Jan 2024

Engineered statistical pipelines to eliminate outliers & enhance data visualization, reducing model complexity by 50%. Analyzed large datasets from multiple sources to identify patterns and trends in data. Designed and built a scalable backend using FastAPI, Pydantic, and Supabase, handling secure document storage, JWT-based authentication, and complex relational data modeling. Automated legal risk assessments by designing system prompts for Gemini to generate structured JSON responses, identifying critical risk flags, key dates, and party roles. Built a dynamic PDF reporting engine using ReportLab to generate formatted executive summaries, complete with risk severity badges and tabular data extraction. Architected a real-time, asynchronous backend using FastAPI and WebSockets, establishing a sub-second communication layer capable of handling high-throughput live stream chat data. Engineered a Multi-Agent AI Pipeline (Moderation, Sentiment, Revenue, and Response generation) powered by Google Gemini, consolidating operations into a single optimized LLM call to reduce latency and minimize API costs. Developed an Intelligent Priority Queue algorithm and Voice Matching Engine to dynamically filter, sort, and auto-advance "super chats" based on sentiment and revenue impact, significantly improving streamer engagement efficiency. Engineered a real-time NLP inference pipeline using FastAPI and HuggingFace Transformers (DistilBERT/GoEmotions), classifying text sentiment and extracting granular emotional distributions with millisecond latency. Built an aspect-based sentiment analysis engine using spaCy to dynamically detect product features from unstructured reviews, providing highly targeted business insights.