Python·PyTorch·TensorFlow·FastAPI·LLM Research·Computer Vision·MLOps·React·Docker·HuggingFace·CGPA 9.15·JSSATE Bengaluru·Python·PyTorch·TensorFlow·FastAPI·LLM Research·Computer Vision·MLOps·React·Docker·HuggingFace·CGPA 9.15·JSSATE Bengaluru·
000 / About

The engineer
behind the code.

I'm Akash G Patil, a 3rd-year B.Tech AI/ML student at JSSATE Bengaluru with a CGPA of 9.15. I build AI systems that go beyond prototypes — focused on reliability, scalability, and real-world deployment.

From detecting hallucinations in LLMs to building smart traffic systems for Bengaluru — I work on problems where accuracy and speed genuinely matter. Currently seeking AI/ML internships and research opportunities.

9.15
CGPA
Top of class
8+
Hackathons
Competed & won
5+
Projects
Shipped to prod
2027
Graduating
Available for Internships and Opportunities
005 / Selected Work

Things I've built.

Each project solves a real problem. No toy demos — built fast, deployed, and production-ready.

001RESEARCH2025

HalluciSense

LLM Reliability Framework

Confidence-aware hybrid framework for detecting and quantifying hallucinations in LLMs. Three-pillar architecture combining retrieval-based verification, model-intrinsic confidence scoring, and cross-consistency checking with a novel H-Score metric.

94%
Detection accuracy
Architecture pillars
H
Score metric
PyTorchHuggingFaceFastAPIResearch
002DEPLOYED2025

GreenWave AI

Smart Traffic Preemption System

Real-time emergency vehicle green corridor system with AI-driven signal sequencing and traffic-weighted Dijkstra routing. Integrated with BBMP traffic signals. Reduces ambulance response time by up to 500ms per intersection.

500ms
Saved per intersection
Live
Deployed on Vercel
BBMP
Signal integration
Next.jsWebSocketsDijkstraBBMP
003BUILT2024

Fraud Detection System

Real-Time ML Pipeline

End-to-end ML pipeline for transaction fraud detection using ensemble methods (Random Forest + XGBoost). Trained on 284k transactions with 99.9% class imbalance. Achieves 97% precision with under 0.1% false positive rate in production.

97%
Precision score
284k
Transactions trained
<0.1%
False positive rate
Scikit-learnXGBoostPythonDocker
004BUILT2024

F1 Race Analytics

Predictive Strategy Engine

ML-powered Formula 1 race strategy system using real telemetry data from FastF1 API. Predicts optimal pit stop windows with 78% accuracy. Built interactive dashboards for lap-by-lap performance visualization.

78%
Pit stop accuracy
22
Drivers modelled
API
Live telemetry
PythonTensorFlowFastF1Pandas
More projects on GitHub
006 / Journey

The journey.

From first model to production systems — the milestones that shaped the engineer.

Education
Project
Hackathon
Research
EDU2021

10th Grade — 95.8%

Kendriya Vidyalaya · CBSE

Strong foundation in Mathematics and Science. Early interest in computers and problem-solving.

2023EDU

12th Grade — 92.5%

Alva's Education Foundation · PCMB

Physics, Chemistry, Mathematics with Biology. Solidified decision to pursue AI/ML engineering.

EDU2023

Started B.Tech — AI/ML

JSSATE Bengaluru

Joined Department of Computer Science & Engineering with specialization in Artificial Intelligence and Machine Learning.

2023BUILD

First ML Projects

Academic Projects

Built Fraud Detection System and NLP Question Answering System. Began exploring PyTorch and HuggingFace transformers.

BUILD2024

F1 Race Analytics + ML Pipeline

Personal Projects

Built Formula 1 race strategy prediction system and an end-to-end ML model deployment pipeline with Docker and FastAPI.

2024HACK

Hackathon Streak Begins

Multiple Competitions

Competed in 8+ hackathons. Developed personal winning formula: idea clarity over code volume, deploy early, invest in presentation.

RESEARCH2025

HalluciSense — Research Project

Major Project · JSSATE

Designed confidence-aware hybrid framework for LLM hallucination detection. Three-pillar architecture with H-Score metric. Target: Elsevier / Frontiers in AI.

2025HACK

GreenWave AI — Build for Bengaluru 2.0

Hackathon · AI/ML Track

Built real-time emergency vehicle green corridor system with Dijkstra routing + WebSocket architecture. Deployed live to Vercel.

EDU2025

CGPA 9.15 · 3rd Year

JSSATE Bengaluru

Maintaining strong academic record while shipping real-world AI projects and competing in hackathons.

007 / Research

In progress.

IN PROGRESSTARGET: ELSEVIER / FRONTIERS IN AI2025

HalluciSense: A Confidence-Aware Hybrid Framework for Detecting and Quantifying Hallucinations in Large Language Models

Akash G Patil, Chirag O, Darshan A, Keerthan B M · JSSATE Bengaluru

We present HalluciSense, a three-pillar hybrid architecture for hallucination detection in LLMs combining retrieval-based verification, model-intrinsic confidence scoring, and cross-consistency checking. Our proposed H-Score metric provides a unified quantitative measure of hallucination severity across generative model outputs, enabling more reliable deployment of LLMs in high-stakes environments.

01
Retrieval Verification
Cross-checks outputs against ground-truth knowledge sources
02
Confidence Scoring
Model-intrinsic uncertainty quantification per token
03
Cross-Consistency
Multi-sample consistency checking for factual stability
LLM HallucinationConfidence ScoringNLPPyTorchHuggingFaceH-Score Metric