Prajwal
Hiremath
Building scalable, production-grade AI systems. A hands-on problem solver passionate about architecting end-to-end solutions—from raw data pipelines to user-facing APIs.
About
Data Scientist & AI Engineer with 5 years of experience building scalable, production-grade AI systems. Proven track record of delivering high-ROI platforms for Pharmaceutical and Enterprise Networking giants, consistently bridging the gap between complex ML research and reliable business applications.
Work Experience
Data Scientist (Consultant)
Deloitte USI
Jan 2025 - Present · Bengaluru, India
OmniScout Multi-Tenant AI Platform
- Architected "OmniScout," a scalable multi-tenant platform consolidating 5+ domain apps into a unified FastAPI system; recognized by leadership for delivering a projected $50M annual benefit and increasing experimental success rates by 30% for 500+ researchers
- Engineered a production-grade Multi-Agent RAG system using LangGraph and AWS Neptune to orchestrate 5+ specialized agents (Query Analyzer, Tool Orchestrator); implemented streaming responses via SSE with automated inline citations and fact-checking
- Developed a Deep Research Agent and SWOT Analysis workflow for pharmaceutical intelligence, utilizing parallel execution to slash research cycles from hours to minutes while maintaining strict scientific accuracy
- Developed a custom MCP server using FastAPI-MCP to standardize and expose internal tools (such as HR, DR, and SWOT analytics) as reusable AI capabilities; streamlined tool discovery and cut cross-team deployment cycles by 40% through a unified Model Context Protocol integration
- Designed a YAML-driven prompt orchestration framework and a custom Shifted Context Window algorithm for query optimization; enabled independent tenant workflows and dynamic configuration loading while eliminating code redundancy across teams
- Developed an end-to-end document intelligence pipeline using Mistral OCR for structured PDF extraction and a custom hybrid search engine (BM25 + Neural Sparse encoding) on OpenSearch; eliminated ML-node dependency by running sparse inference externally, achieving high-precision retrieval across pharmaceutical research documents.
Eli Lilly Chat Agent
- Architected an advanced SQL Agent using LangChain and Anthropic Claude on AWS Bedrock, enabling natural language querying of AWS RDS; achieved 95% query accuracy, outperforming OpenAI benchmarks and securing client retention
- Engineered a dynamic few-shot learning system (OpenSearch) and Chain of Thought (CoT) prompting pipeline; optimized logic for complex joins and validation, boosting domain accuracy by 18% and reducing hallucinations by 31%
- Developed a Multi-Agent RAG ecosystem using CrewAI to orchestrate 4 specialized agents (Router, Retriever, Synthesizer, Evaluator) for processing complex clinical trial documents and scientific data
- Implemented a custom hybrid search engine combining FAISS vector similarity with BM25 keyword matching and DeepEval for automated quality metrics, ensuring high-precision retrieval for pharmaceutical research
Software Engineer
Cisco Systems India Pvt Ltd
Jan 2021 - Dec 2024 · Bengaluru, India
Testbed Reservation System
- Architected a high-availability Django platform scaled to serve 300+ engineers and 5,000+ monthly reservations, eliminating 100% of double-booking race conditions by implementing PostgreSQL row-level locking within atomic transactions
- Engineered a non-blocking Celery-Redis asynchronous notification system and built utilization dashboards, which together reduced testbed idle time by 45% (recovering $150K in annual value)
ML Testbed Predictor
- Engineered an ML predictive pipeline to optimize testbed allocation, benchmarking 7+ algorithms; achieved 0.91 AUC using Gradient Boosting, outperforming baseline models by 35% and reducing idle time allocation failures
- Implemented a robust preprocessing workflow using StandardScaler and LabelEncoder, serializing the pipeline with joblib to prevent training-serving skew; served real-time predictions via a Flask REST API
Cisco Wiki Analytics Dashboard
- Engineered a Flask and Pandas dashboard to automate executive reporting by ingesting, cleaning, and programmatically standardizing raw Excel data, which reduced manual report generation time by 40%
- Delivered hierarchical, interactive chart views tailored for leadership, enabling Directors to see high-level aggregates and Managers to drill-down into specific team/engineer performance for data-driven decision-making
Testbed Chatbot
- Built a RAG chatbot (FastAPI, OpenAI Embeddings, ChromaDB) that enabled natural language queries for hardware specs, reducing information lookup time by 75%
Personal Projects
Sarcasm Detection with BERT
NLPBuilt a sarcasm detector using custom classification layers on frozen BERT embeddings with batch tokenization via collate_fn.
- Custom layers on frozen BERT backbone
- Batch tokenization with dynamic padding
- Binary classification with high precision
Image Classification & Transfer Learning
VisionBuilt a multi-layer CNN with BatchNorm/Dropout for image classification and fine-tuned ResNet-18 by freezing backbone layers.
- Custom CNN with BatchNorm and Dropout
- ResNet-18 fine-tuning with frozen backbone
- Stratified validation splits
Audio Classification with Mel Spectrograms
Audio MLDesigned a 2D-CNN architecture for audio classification with on-the-fly Mel Spectrogram generation from raw audio.
- 2D-CNN for spectrogram classification
- On-the-fly Mel Spectrogram generation
- Robust audio preprocessing pipeline
Tabular Data Classification
MLBuilt a deep neural network for binary classification on tabular data with sklearn preprocessing pipelines.
- DNN with sklearn preprocessing
- Feature engineering pipeline
- Serialized inference with joblib
Technical Skills
Programming & Languages
Agentic AI & Frameworks
GenAI & LLMs
Data & Vector Databases
Web & Infrastructure
Deep Learning & ML
Certifications
NVIDIA Certified Associate: Generative AI and LLMs
NVIDIA
AWS Certified AI Practitioner
Amazon
Databricks Certified Generative AI Engineer Associate
Databricks
Google Cloud Generative AI Leader (GCP-GAIL)
Azure AI Engineer Associate (AI-102)
Microsoft
Azure AI Fundamentals (AI-900)
Microsoft
CrewAI Multi Agent System
DeepLearning.AI
Programming Expert
ProgrammingExpert.io
Google IT Automation with Python
Coursera
Introduction to Generative AI and LLMs
Python for Data Science and ML Bootcamp
Udemy
The Complete SQL Bootcamp
Udemy
Latest from Medium
Key Achievements
OmniScout recognized by leadership for delivering a projected $50M annual benefit and increasing experimental success rates by 30% for 500+ researchers
Secured $2M+ pharmaceutical client engagement with 13% accuracy improvement over competitor LLMs
Built 5 production deep learning models across tabular, vision, audio, and NLP domains
Reduced operational time by 40-80% through ML-powered automation and data-driven insights
Education
Bachelor of Technology in Electrical and Electronics Engineering
Jawaharlal Nehru Technological University, Anantapur
CGPA: 8.11