Building AI Systems
for Real-World Impact.
Designing, deploying, and scaling machine learning and AI systems across real-world applications.
Role & Positioning
I am focused on the rare intersection of Scientific Machine Learning (SciML) and Production AI Systems.
My expertise lies in Physics-informed ML, thermodynamic optimization, and uncertainty-aware modeling to map natural phenomena into actionable mathematical spaces.
To execute these models at scale, I build robust systems: highly optimized LLM pipelines (RAG, SLMs, alignment), ultra-low latency real-time computer vision networks, and scalable ML infrastructure.
Systems Engineering
Engineering Projects
EcoGuard — AI Carbon Intelligence
Problem
Need for precise multi-modal intelligent tracking for carbon usage linked dynamically with systems.
Solution
Built a multimodal system integrating IoT, Computer Vision, and Ensemble ML providing carbon scores, breakdown, real-time tracking, and what-if simulations.
Vision System for Physical AI
Problem
Robotic cleaning systems require extremely robust multi-class debris detection capabilities at the edge.
Solution
Engineered a YOLOv8 Nano-based perception system with a custom self-built dataset prioritizing edge-efficiency.
Institutional Language Model
Problem
Healthcare domains require strictly factual language generation without external API telemetry.
Solution
Developed an On-prem Healthcare SLM + RAG system incorporating a Vector DB alongside strict role-based access controls.
AUDITRON — AI Accounting Agent
Problem
Auditing mechanisms for invoices and ledgers were highly manual and prone to visual reasoning errors.
Solution
Fine-tuned a Donut VLM to automate invoice and ledger validation within a secure inference pipeline.
Multi-Agent RAG System
Problem
Complex queries over massive document datasets returned severely degraded vector search results.
Solution
Implemented a multi-agent orchestrated reasoning pipeline querying over 100k+ documents simultaneously.
Publications & Modeling
Research & Papers
Carnallite Electrolysis Optimization
Problem
Industrial Mg electrolysis operates at 650–720°C, dictated by electrolyte liquidus. This requires intense thermal constraints.
Method
Developed Physics-Informed Gaussian Process (PI-GP) model utilizing Differential Evolution with physical constraints. Built 134-point dual-source dataset (FactSage + NIST).
Key Results
Derived Eutectic Lower Bound Theorem establishing 423°C absolute minimum. Identified optimal MgCl₂–NaCl eutectic pathway at 459±1°C. R² = 0.9996 / RMSE = 1.5°C within-source.
Magnesium Extraction Optimization
Problem
Extraction processes required highly inefficient extreme temperature operations.
Method
Physics-informed ML + thermodynamic modeling combining CALPHAD and FactSage cross-validation.
Key Results
Introduced Slag Basicity Extractability Theorem. Achieved 250°C reduction (1200°C → 800°C) with Gradient Boosting modeling (R² = 0.987).
Medical LLM Structural Alignment
Problem
Current health AI reasoning architectures misalign structure logic against pure factual queries. Accuracy ≠ deployment readiness.
Method
Identified the Structural Alignment Gap mathematically and introduced the Semantic Drift Score (SDS) and Structural Alignment Index (SAI).
Key Results
Categorized logic flows establishing concrete guardrails for clinical environments.
Levulinic Acid Yield Prediction
Problem
Predicting chemical yields across complex non-linear catalysts.
Method
Benchmarked 10 distinct models across a highly engineered 200-point physical dataset. Identified GPR as optimal.
Key Results
Achieved R² = 0.966, formally isolating the optimal process region for sustainable yields.
Recursive Language Models (RLM)
Problem
Language models lack true environmental reasoning limits during long-context execution.
Method
Architected long-context scaling via targeted semantic recursion loops based on environmental parameters.
Key Results
Defined new efficiency bounds for inference-time reasoning architectures.
Renewable Energy Forecasting
Problem
Stochastic renewable energy outputs disrupted broader power grid intelligence.
Method
Constructed an Ensemble ML framework benchmarking live output data to link ML performance to direct grid efficiency.
Key Results
Forecast errors reduced by ~30% on synthetics and ~15% in complex real-world trials. Best Research Paper Award — ICCTVB-25.
Professional History
Career Timeline
AI Developer Intern
Engineered production LLM and RAG systems achieving 35% retrieval accuracy improvement and 45% latency reduction.
Architected Dockerized inference pipelines with CI/CD and canary deployments, cutting deployment time to under 10 minutes.
Designed scalable inference services with cost optimization through quantization and post-deployment monitoring.
AI & ML Intern
Delivered 5+ NLP applications using AWS Lambda and Streamlit, reducing manual processing time by 30%.
Automated text summarization and classification workflows, increasing operational throughput by 25%.
Capabilities
Technical Stack
ML / AI
LLM Systems
MLOps
Tools
Core Expertise
Milestones
Achievements & Accolades
Achievements
Code4Society Hackathon (2026)
International Conference on Convergence of Technologies for Viksit Bharat (ICCTVB-25), Sanjay Ghodawat University, Kolhapur (Scopus Indexed) 2025
National DevCraft Hackathon (Team Tech Titans), Fluxus 2025, IIT Indore 2025
National AI Hackathon, IIT Indore 2025
Table Tennis Men’s Singles, Atharv Ranbhoomi 2024, IIM Indore 2024
Table Tennis Doubles Competition, IIT Indore 2024
Table Tennis Men’s Singles, Pimpri Chinchwad University 2023
Table Tennis Tournament, Kridarambh 2K24, Pimpri Chinchwad University 2024
2 Inter-College Coding Competitions 2024
Delivered 3+ technical talks on AI deployment, ML ethics, and production ML systems
Deep Learning Specialization Certificate, Deep Learning.AI (Coursera) 2025
Supervised Machine Learning Certificate, Stanford University (Coursera) 2025
Mathematics for Machine Learning and Data Science Certificate, Deep Learning.AI (Coursera) 2025
Key Courses Taken
Deep Learning
Deep Learning Specialization (DeepLearning.AI), TensorFlow for AI/ML (DeepLearning.AI)
Machine Learning
Supervised ML (Stanford University), ML with Python (IBM)
Mathematics
Mathematics for ML & Data Science - Linear Algebra, Calculus, Probability & Statistics (DeepLearning.AI)
Positions of Responsibility
Core Coordinator
PCU Sports Club and Committee, Pimpri Chinchwad University, Pune 2023-27
Placement Coordinator
Career Development Cell, Pimpri Chinchwad University, Pune 2024-27
Core Team Member
Research & Development Club, Pimpri Chinchwad University, Pune 2023-27
Team Leader
Team Arogya codes, Smart India Hackathon 2025 (Selected at PCU Internal Hackathon) 2025
Sports Coordinator
Kridarambh 2K24, Pimpri Chinchwad University, Pune 2024
Coordinator
Project Expo 2K24, Pimpri Chinchwad University, Pune 2024
My Gallery
A visual log of hackathons, lab physical experiments, and deployments.
.jpeg&w=3840&q=75)
.jpeg&w=3840&q=75)
.jpeg&w=3840&q=75)
.jpeg&w=3840&q=75)
