Hello, I'm

Akhila
Susarla

AI/ML Engineer • Data Scientist

Building production-grade conversational AI systems, ML pipelines, and data-driven solutions. Currently engineering voice bots handling 10,000+ daily interactions at Copart.

LinkedInLinkedIn GitHubGitHub

3+

Years Experience

<2s

Response Latency

LoRA

LLM Fine-tuning

3.78

GPA at UT Dallas

What I Do

Conversational AI & Voice Bots

Building production-grade voice AI systems with real-time speech processing, IVR navigation, and natural conversation flow optimization.

LiveKitDeepgramElevenLabsGeminiGPT

ML Pipeline Development

End-to-end machine learning pipelines from data preprocessing to model deployment, with MLOps automation and continuous monitoring.

scikit-learnPyTorchTensorFlowKubernetesSpinnaker

Data Science & Analytics

Extracting actionable insights from large-scale datasets through statistical modeling, EDA, time-series forecasting, and anomaly detection.

PandasPySparkHadoopDatabricksBigQuery

NLP & LLM Engineering

Fine-tuning and deploying large language models with prompt engineering, RAG pipelines, and conversational AI applications.

LangChainLoRA/QLoRATransformersBERTNER

Skills & Tools

Languages

AI & GenAI

Data Science & ML

Tools & Frameworks

Cloud & DevOps

PythonSQLTypeScriptJavaScriptC/C++

LLMs (Gemini, GPT, Claude)LangChainPrompt EngineeringLiveKitDeepgram (STT)ElevenLabs (TTS)LoRA / QLoRARAG

NumPyPandasscikit-learnPyTorchTensorFlowPySparkHadoopStatistical ModelingEDATime-Series ForecastingClustering & SegmentationAnomaly DetectionModel EvaluationCross-ValidationNLPLDAPCA

GitJupyterDatabricksDatadogPostmanNode.jsGraphQLMongoDBMariaDBMS SQL ServerApache KafkaElasticsearchn8nExcel

AWS (SageMaker, S3, Lambda)Azure DevOpsKubernetesSpinnakerCI/CDGCP (Vertex AI, BigQuery)Docker

Experience

February 2026 - Present

Dallas, TX

June 2025 - August 2025

Fort Worth, TX

July 2023 - July 2024

Pune, India

July 2022 - June 2023

Pune, India

June 2021 - August 2021

Amaravati, India

Projects

Real-time NER Radar

Deployed

Built a large-scale real-time data pipeline processing Reddit comments through Apache Kafka and PySpark to extract named entities and visualize trending topics in Kibana, demonstrating expertise in distributed stream processing and NLP at scale.

Real-time StreamingNER at ScaleKibana Dashboards

Apache KafkaPySparkNLTKElasticsearchKibanaLogstashReddit API

End-to-end streaming pipeline: Reddit API → Kafka → PySpark → Elasticsearch → Kibana
NLTK-based NER extraction surfacing entity frequency trends in real-time dashboards
Captured and analyzed data at 15/30/45/60-minute intervals over a 1.5-hour run

GitHub →

Influence Maximization in Complex Networks

Published

Co-authored a Springer-published research paper presenting algorithmic solutions for identifying the most influential nodes in complex networks — a core challenge in viral marketing, epidemic control, and social network analysis.

Springer PublishedFICTA 2023NP-Hard Optimization

PythonGraph TheoryNetwork AnalysisGreedy Algorithms

Solved the NP-hard seed selection problem for maximizing information spread across heterogeneous networks
Benchmarked greedy and heuristic approaches against centrality-based methods on real-world social graph datasets
Demonstrated scalable influence prediction applicable to viral marketing, public health campaigns, and recommendation systems

Research Paper →

CorroSight

Deployed

Built an AI-powered pipeline integrity platform that automates alignment of In-Line Inspection data across 15+ years and multiple vendors, predicting corrosion degradation and generating prioritized repair schedules from fragmented datasets.

Digital Twin in <10s59+ Event TypesAI Risk Simulation

PythonFastAPIAngularxAI GrokHungarian AlgorithmNLP

Processes raw inspection data into functional digital twins in under 10 seconds
Hungarian Algorithm-based optimal defect matching with 1,603 ground-truth anchors
Integrated xAI Grok for AI-driven risk simulation and natural language querying

Devpost →

UniWay — Campus Navigation & Analytics

Prototype

Designed and prototyped an intelligent campus navigation and space-booking platform solving real scheduling conflicts through real-time room availability tracking and interactive campus mapping, built in a 24-hour hackathon.

2nd Runner-UpCBRE Partnership24hr Hackathon

FigmaAnalyticsUI/UX DesignProduct Strategy

Secured 2nd Runner-Up at UXperience The First Draft hackathon for CBRE
Real-time room availability tracking and interactive campus mapping for hundreds of daily users
Combined engineering rigor with intuitive UX — learned Figma from scratch in the hackathon

Figma →Devpost →

Sentiment Analysis on Social Media Data

Completed

Developed a multi-model sentiment analysis framework comparing TF-IDF, Word2Vec, GloVe, and LSTM approaches against fine-tuned BERT, achieving 81.56% accuracy on US airline tweet sentiment classification.

81.56% AccuracyBERT Fine-tuned5 Model Comparison

PythonBERTLSTMTF-IDFWord2VecGloVescikit-learn

Rigorous comparison across 5 embedding/model architectures for production-grade sentiment systems
Fine-tuned BERT achieving 81.56% accuracy on multi-class airline sentiment classification
End-to-end NLP pipeline: data preprocessing, tokenization, feature extraction, and model evaluation

GitHub →

Mushroom Species Classifier

Completed

Built a deep learning image classifier using transfer learning with EfficientNetB0 to identify 9 mushroom genera from ~6,700 web-scraped images, tackling real-world class imbalance and noisy data through a two-phase training strategy.

9 Genera6,700 ImagesEfficientNetB0

PythonTensorFlowKerasEfficientNetB0Transfer Learning

Two-phase training: frozen feature extraction then fine-tuning with partially unfrozen layers
Advanced augmentation pipeline (rotation, shifts, zoom, brightness) to handle class imbalance (311–1,563 samples/class)
Custom classification head with dropout, batch normalization, and learning rate scheduling

GitHub →