Chirudeva Reddy // AI Engineer
01 / What I do
AI engineer and 3rd-year Computer Science undergraduate at BITS Pilani Dubai, with a Minor in Data Science.
02 / Research
Estimating parametric 3D human body meshes from 2D binary silhouettes, without ever storing raw imagery. An IJCAI paper proposal under Dr. Shivang Agarwal.
03 / Industry
−42% API error rate at Nablix Labs. +28% efficiency on fullstack ML apps at Jovens Technologies. Security automation across 8 client platforms at Domdog.
My academic research focuses on the intersection of Computer Vision, Deformable 3D Representations, and Privacy-Preserving Perception. Under the supervision of Dr. Shivang Agarwal, I am authoring a paper proposal for IJCAI on estimating parametric 3D human body meshes directly from 2D binary silhouettes without storing raw optical imagery.
In industry, I work as AI Engineer Intern at Nablix Labs (curating API endpoints and reducing error rates by 42%), and previously served as Fullstack Developer Intern at Jovens Technologies FZ-LLC (architecting fullstack ML apps with 28% efficiency boost and NLP pipelines reducing manual time by 12%), and SDE Intern at Domdog (building Playwright automation, malicious content scanner integration across 8 client platforms, and boosting SEO traffic by 11%).
The technologies and tools I use to build, train, analyze, and ship intelligent systems at scale.
Python, SQL, Java, C, C++ and high-performance algorithms.
Transformers, BERT, embeddings, semantic search and clustering.
Cloud storage, data pipelines, Docker, and scalable services.
Perception, segmentation, tracking, pose estimation, and 3D reconstruction.
Deep learning, classical ML, model evaluation & optimization.
FastAPI, microservices, REST APIs, and production-ready systems.
Dashboards, data storytelling, version control, Excel & LaTeX.
Pandas, NumPy, Matplotlib, EDA, preprocessing and visualization.
End-to-end testing, vulnerability scanning, CI/CD, and automation workflows.
Creating a pipeline to estimate human body composition from silhouettes under privacy constraints using transfer learning, hyperparameter tuning, and 3-D reconstruction under supervision of Dr. Shivang Agarwal.
PyTorch, SAM 2.1, YOLOv11, Siamese ResNet-18, SMPL-X
Catching RAG hallucinations in a transformer's hidden states before the first wrong token is written. Paired forward passes on Qwen2.5-1.5B, with and without the retrieved evidence, yield four representation-drift signals fused into one score: AUROC 0.651 on RAGTruth and 0.645 zero-shot on HaluEval, with drift peaking two tokens before the hallucination appears.
PyTorch, Qwen2.5-1.5B, hidden-state probing, RAGTruth
Built a collaborative 4-member project on crash severity prediction for the Chicago Crash Dataset of about 940,000 records. Contributed data-preprocessing and interactive risk visualization.
LightGBM, SMOTE, SHAP / LIME, DBSCAN, GeoPandas
Worked under Dr. Shanookha Ali to build a system using fuzzy logic and artificial neural networks in order to tune and predict continuous depression risk of students from lifestyle and academic factors, achieving 98% training error reduction.
ANFIS, Sugeno FIS, nested cross-validation, calibration
Four roles, from production APIs to the computer architecture lab. One through-line: turning complex systems into measurable, dependable outcomes.
