Joel Markapudi

AI/ML Engineer | Data & BI Engineering | M.S. in Artificial Intelligence @ Northeastern University

AI/ML engineer finishing an M.S. in Artificial Intelligence at Northeastern University (GPA 3.9), with industry experience across AI engineering solutions, financial ML, data platforms, BI and reporting. These days I build retrieval and agent systems that hold up on real data: document intelligence, RAG with traceable citations, multimodal retrieval, multi-agent LLM systems, and LangGraph orchestration with evaluation built in, plus computer vision projects from my M.S. I care most about the parts that make a system trustworthy: clean data, measured cost, and results that reproduce.

Research & Academic Projects

Click on each project card to explore more details!

FinSights: Production-Grade Financial Intelligence System
Hybrid RAG over SEC 10-K filings: structured KPI lookup plus semantic search across 614K sentence embeddings, streamed from AWS Fargate at $0.014 per query with answers cited to exact filing sentences.
Text-to-Pose: CLIP-Conditioned Diffusion for 3D Human Pose
CLIP-conditioned diffusion model that generates 3D human skeletons from captions. A retrieval evaluator reproduces the published R-Precision of 0.797, and a conditioning analysis looks at how CLIP handles left and right.
SocrAItic Circle: Multi-Agent LLM Debate & Evaluation
Multi-agent LLM debate framework with judges built on argumentation theory, rebuilt as a local LangGraph and MCP system. Hybrid retrieval grounds the evidence, and failure-mode instrumentation caught nine silent defects.
Multi-View 3D Scene Analysis
10k+ line pipeline for pose-guided multi-view feature correspondence on ETH3D, with RANSAC validation, occlusion-aware multi-view tracks and a COLMAP mapper integration.
Protein Structure Prediction
HMM, CRF and BiLSTM models for protein secondary structure prediction on the CB513 benchmark. The CRF reached 67% accuracy using evolutionary features and specialized β-sheet features.
Vision-Language Late-Interaction Retrieval
ColBERT-style retrieval that pairs ViT patch tokens with CLIP text tokens through a learned projection, fine-tuned with LoRA on 1,500 Flickr8k pairs. MaxSim scoring with symmetric InfoNCE cut the training loss by 38%.
Artist Classification
Comparison of SVM-SIFT-BoVW, autoencoders and CNNs for artistic style classification across 50 artists. The classical SVM reached 89.2% accuracy, ahead of ResNet-50 and VGG-16.

Professional Experience

Data Platform & Analytics Engineering
Built the enterprise data warehouse (40+ star-schema tables) and entity resolution across six systems at 88% accuracy. Delivered 25+ Power BI dashboards with 300+ KPIs, saving 70+ hours a month, and cut deployment time 50% with CI/CD.
Financial Machine Learning
Production ML for finance and operations: an attrition hazard model that lifted AUC from 0.56 to 0.71, rate guidance, collections and payment-timing models, and anomaly detection that surfaced $37.17M in revenue against near-zero pay.

Other Works

Additional Labs & Experiments

OSCD Sentinel-2 Satellite Change Detection
A side project pursued in depth for learning remote sensing from first principles. A Siamese U-Net is compared against classical baselines and reaches about 1.6x their F1. → Project Page | → Source Code
Airflow, Docker and Qdrant RAG Ingestion Lab
A multi-service Docker Compose stack running a 5-task Airflow DAG into Qdrant, plus a parquet sharding pass that compressed 44GB of files into a 1.5GB, 71M-row table. → Source Code
DNA Pattern Matching: EPMSPP and Ukkonen Suffix Trees
Java implementations of EPMSPP and Ukkonen's suffix tree, benchmarked against KMP, Boyer-Moore, Rabin-Karp and Aho-Corasick on DNA sequences up to 100M base pairs. → Source Code
ML, DL and CV Learning Experiments
A long-running practice repo with from-scratch Transformer reconstruction labs (encoder, decoder-only, encoder-decoder and ViT) in MLX and PyTorch, plus LangGraph agent fundamentals. → Source Code
ML-Serving with GitHub CI/CD and AWS Lambda
An end-to-end ML deployment workflow that uses GitHub Actions, AWS Lambda and SAM for serverless model serving. → Source Code
Optuna and MLFlow with Synthetic Time-Series
Hyperparameter optimization and experiment tracking with Optuna and MLflow on synthetic time-series data. → Source Code