Hi, I’m Aashrit

Aashrit
Bagareddy.

Senior Data Scientist & AI Engineer

I build production ML systems, real-time inference pipelines, and LLM-powered agentic AI, from deep-learning models to multi-agent platforms with their own tooling, RAG, and evals.

View Projects Get in Touch
0 +
Years experience
0 +
Models deployed
$ 0 K+
Revenue driven
$ 0 M
Client savings

01 About

I’m a Senior Data Scientist & AI Engineer with 4 years of experience designing, training, and deploying production ML systems at scale - deep-learning models, real-time inference pipelines, and LLM-powered agentic AI. My work has driven ~$700K in revenue and $2.5M in client cost savings.

At Quantico Energy Solutions I’m the product owner for QLog, a no-code ML platform for upstream oil & gas data science, and I independently built QLogRT - a real-time inference platform that serves ahead-of-bit predictions at the rig site and drove ~$500K in revenue with a major national oil operator. Along the way I extended model prediction range 2.5×, improved accuracy 25%, and managed 100+ model versions end-to-end.

Most recently I exposed the platform’s 12-node DAG workflow engine to LLM agents through a production MCP server and a hybrid code-aware RAG pipeline, and built agents for workflow composition, node scaffolding, and failure triage - hardened with eval-driven development (Inspect AI, Arize Phoenix) and human-in-the-loop approval.

I hold an M.S. from Texas A&M University (3.9 GPA) and certifications in IBM AI Engineering, IBM RAG & Agentic AI, and AWS Cloud Practitioner.

Agentic AI

MCP servers, multi-agent systems (LangGraph + Claude Agent SDK), hybrid code-aware RAG, and eval-driven development.

Real-Time ML

Production inference at the rig site - WITSML streaming ingestion, ahead-of-bit prediction, sub-10-second latency.

Deep Learning

Hybrid Transformer-LSTM ensembles extending prediction range 2.5× with a 25% accuracy improvement over legacy models.

02 Experience

Senior Data Scientist

Quantico Energy Solutions

Jan 2025 - Present
  • Built LLM agents on LangGraph + Claude Agent SDK - a ReAct workflow-composition agent, a node-scaffolding agent, and a multi-agent failure-triage system - with tool sandboxing, workflow-vs-agent boundaries, and human-in-the-loop approval.
  • Exposed the platform’s 12-node DAG engine to LLM agents via a production MCP server (JWT-authed, rate-limited, cost-gated) and a hybrid code-aware RAG pipeline (pgvector + full-text + graph → RRF → cross-encoder rerank).
  • Drove eval-driven development with Inspect AI - custom scorers benchmarking DAG-composition validity at 0.96 and codebase-retrieval recall@10 - wired into CI with Arize Phoenix / OpenTelemetry tracing.
  • Independently built QLogRT, a real-time ahead-of-bit ML platform serving inference at the rig site (~$500K revenue with a major national oil operator).

Data Scientist

Quantico Energy Solutions

Jun 2022 - Jan 2025
  • Product owner for QLog, a no-code ML platform for upstream O&G data science - model building, data processing, visualization, and training across well-log, drilling, and production datasets.
  • Built prescriptive ML models and data-science workflows for an operator, generating $2.5M in operational cost savings and $300K in direct revenue.
  • Productionized hybrid Transformer-LSTM models against legacy architectures - extended prediction range from 30ft to 100ft (2.5×) and improved accuracy 25%.
  • Established end-to-end MLOps: MLflow tracking, automated registration of 100+ ensembles, H3-geospatial model selection, and GitHub Actions CI/CD.

Data Science Intern

Quantico Energy Solutions

Apr 2022 - Jun 2022
  • Data warehousing and ETL with PySpark for energy analytics; streamlined data-processing workflows and prototyped early ML models.

03 Featured Projects

Each project leads with its impact and the full technology stack, followed by the architecture behind it. Expand any diagram for a closer look.

02

QLogRT Real-Time ML Platform

Ahead-of-bit log prediction and drilling optimization at the rig site

Proprietary, Quantico Energy Solutions

A containerized real-time platform that ingests WITSML drilling data, runs LSTM/FFNN ensemble predictions ahead of the bit (DTCO/DTSM/RHOB) plus cluster-based ROP optimization, and streams recommendations to a live dashboard. Consolidated multiple services into a unified async FastAPI backend with retry / circuit-breaker resilience and WebSocket-driven visuals.

~$500Krevenue, national oil operator
<10sstreaming inference latency
2.5xprediction range extended
FastAPITensorFlow / KerasMLflowMongoDBNext.jsDocker ComposeWITSMLSocket.IOPlotly DashSHAP

Architecture · click to expand

Real-Time ML
Rig Data WITSML stream Preprocess time → depth REAL-TIME ML Log Ensemble LSTM / FFNN Drilling Optimizer cluster + NN sim Ahead-of-bit logs DTCO · DTSM · RHOB ROP + setpoints WOB · RPM Live Dashboard at the rig site Model Registry MLflow · geo-select

Rig data is preprocessed, run through log-prediction and drilling-optimization models, and streamed to a live dashboard at the rig.

03

Medical CV, 3D Organ Segmentation

Specialist vs. foundation model on CT/MRI volumes

GitHub Live Demo

End-to-end 3D medical-imaging pipeline: train volumetric SegResNet networks with MONAI, serve them on NVIDIA Triton behind a FastAPI pre/post layer, and benchmark against NVIDIA’s VISTA3D foundation model, explored in a browser NiiVue viewer. Trained 10 Medical Segmentation Decathlon tasks across CT and MRI.

0.95val Dice (liver / spleen)
10MSD tasks trained
TritonGPU inference serving
PyTorchMONAISegResNetVISTA3DNVIDIA TritonFastAPINiiVueTorchScriptDocker

Architecture · click to expand

Segmentation
CT / MRI MSD dataset 3D Preprocess MONAI SEGMENTATION SegResNet trained specialist VISTA3D foundation model served on NVIDIA Triton Segmentation masks Dice benchmark Browser Viewer NiiVue 3D

CT/MRI volumes are segmented by a trained SegResNet and NVIDIA's VISTA3D foundation model, then compared by Dice in a browser viewer.

04

Deep Learning Ensembles

Hybrid Transformer-LSTM for ahead-of-bit well-log prediction

Proprietary, Quantico Energy Solutions

20+ member ensembles combining Transformer and LSTM/TCN architectures for streaming well-log prediction on WITSML data. Extended prediction range from 30-40ft to 100ft and improved accuracy 25% over legacy architectures, with conformal uncertainty and H3-geospatial model selection.

2.5xrange (30 to 100 ft)
+25%accuracy vs. legacy
100+registered model versions
PyTorchTensorFlowLSTMTCNTransformersEnsemble LearningConformal PredictionMLflow

Architecture · click to expand

Ensemble Model
Log window WITSML stream HYBRID MODEL Transformer attention LSTM / TCN sequence Fusion head Ensemble 20+ members Ahead-of-bit 30 → 100 ft Uncertainty conformal band

A streaming log window feeds a hybrid Transformer + LSTM/TCN ensemble that predicts logs ahead of the bit with a conformal uncertainty band.

04 Skills & Tools

LLMs / GenAI / Agentic AI

Claude Agent SDKMCPLangGraphLangChainPydantic AIMulti-Agent SystemsReAct Tool-UseHybrid RAGCross-Encoder Rerankingpgvector / pgvectorscaleChromaDBFAISSHugging Face TEIBGE-M3Qwen3-EmbeddingPrompt Engineering

AI / ML

PyTorchTensorFlowscikit-learnXGBoostTransformersLSTMTCNCNNFoundation ModelsTime-Series ForecastingEnsemble LearningNAS / OptunaConformal Prediction

MLOps / Eval / Observability

MLflowInspect AIpromptfooEval-Driven DevelopmentArize PhoenixOpenTelemetry GenAIDockerKubernetesGitHub Actions CI/CDAWSAzure

Languages / Data / APIs

PythonTypeScriptNode.jsSQLFastAPIPostgreSQLMongoDBPySparkPandasNumPyPlotly DashStreamlitWebSocketsMicroservices

05 Education & Certifications

Texas A&M University

M.S. Petroleum Engineering

College Station, TX · 3.9 GPA · 2021

Birla Institute of Technology

B.E. Mechanical Engineering

India · 2018

Certifications

  • IBM AI Engineering Professional
  • IBM RAG & Agentic AI Professional
  • AWS Cloud Practitioner (CCP)

What’s next

Let’s build something
together.

I’m open to conversations about AI/ML engineering, agentic systems, and applied research: new roles, interesting problems, or just comparing notes with fellow engineers.

aashritbagareddy@gmail.com
LinkedIn GitHub +1 (512) 423-7198