Building intelligent systems that combine machine learning, optimization, and real-world impact.

• 🎓 B.Tech CSE, KIIT University (2026)
• 🤖 Focused on Applied AI Systems, LLM Fine-Tuning, & NLP
• 📊 Strong in Data Modeling, EDA, and Predictive Analytics
• ⚡ Interested in building high-performance, edge-deployable ML systems
• 🧠 Passionate about combining technology + leadership + product thinking

Large Language Models (LLMs), Parameter-Efficient Fine-Tuning (PEFT, QLoRA)
Hugging Face Ecosystem (Transformers, TRL, BitsAndBytes)
TF-IDF, BART, Text Summarization, Classification, Ensemble Methods

Python (Pandas, Scikit-Learn, Flask), SQL, Data Cleaning, Feature Engineering

Looker Studio, Power BI, Excel, Exploratory Data Analysis (EDA), Trend Analysis

Relational & Dimensional Data Modeling, Schema Analysis, Query Optimization
SAP Business Data Cloud, Supabase

Tech: Python, Hugging Face (PEFT, TRL), QLoRA, TinyLlama, Google Colab

• Fine-tuned a Small Language Model (TinyLlama-1.1B) using QLoRA (4-bit quantization) to strictly categorize unstructured customer support tickets into 77 distinct banking intents.
• Utilized Hugging Face trl and peft libraries to train low-rank adapters, reducing trainable parameters by over 99% for highly efficient deployment.
• Designed an automated inference pipeline for zero-shot text classification, transforming messy qualitative feedback into structured quantitative metrics for product analytics.

Tech: Looker Studio, Python (Pandas), Google Sheets

• Architected an end-to-end analytics pipeline analyzing 1,000+ retail leads to identify sales bottlenecks.
• Uncovered a critical 18.8% funnel drop-off at the pre-booking stage and calculated stage-to-stage conversion rates.
• Discovered an 8.3% revenue leakage and a severe post-sales service deficit (average NPS of -53).

Tech: React Native, Flask, Machine Learning, ESP32

• Built an end-to-end IoT-enabled precision agriculture system.
• Developed an ML crop recommendation engine with 87% accuracy.
• Integrated real-time soil and weather data for multi-parameter analysis.

Tech: Python, NLP, Schema Optimization

• Designed an NLP system converting natural language to optimized SQL.
• Improved complex query execution speed by 30–35%.
• Implemented schema-aware query optimization for enhanced database retrieval.

Qualifying Predictor

• Built a telemetry-based ML model achieving 82% prediction accuracy for race outcomes.

Hybrid Report Summarizer

• TF-IDF + BART-based summarization pipeline.
• Reduced manual document review time by 60%.

📧 ayushkedia.er@gmail.com
🔗 LinkedIn: linkedin.com/in/ayush-kedia
🎓 KIIT University (2026)