I amPremithChilukuri

ABB

AI Engineer

2024 — PRESENT

AI Engineer · Mar 2025–present / AI Intern · Jan–Mar 2025 / Robotics Intern · May–Dec 2024

Building the infrastructure that turns agentic AI into dependable enterprise systems.

1. 01

   Enterprise agent orchestration

   Architected a natural-language automation platform with a hierarchical supervisor–specialist runtime, parallel DAG execution, stateless container workers, and Kubernetes scaling.

2. 02

   Knowledge at industrial scale

   Built an agentic multimodal RAG platform spanning a 400K+ document corpus, a 419K-entity knowledge graph, and 1.66M relations, using PostgreSQL, Apache AGE, pgvector, query expansion, and relevance grading.

3. 03

   Long-running agents, under control

   Engineered persisted human approval checkpoints, automatic context compression, token budgets, dynamic skill loading, sandboxed Python, and end-to-end tool-call tracing.

4. 04

   Connected tools and usable outputs

   Implemented MCP integrations, OAuth connectors, structured SAP analytics, and versioned document generation. Served the platform through FastAPI and deployed it in Docker with multi-provider inference.

5. 05

   Reusable agent capabilities

   Built a dynamic skills framework for agents to author, version, and selectively load task-specific capabilities, with sandboxed execution and structured input/output contracts.

6. 06

   Research into engineering practice

   Contributed to MiRAGE and Power Circuit AI. Earlier ABB work covered evaluation across 500+ queries with 10+ metrics, motor search over 500+ specifications, and a 6D pose pipeline reporting 5 mm MSSD at 0.6 s inference. Internship contributions also included LLM-based product filtering and evaluation, vision-language 2D-to-3D CAD prototyping, and language-guided object isolation with Phi-3 and SAM2.

Virginia Tech

Graduate Research Assistant

2023 — 2024

March 2023 – May 2024

Connecting generative modeling, manufacturing constraints, and applied machine learning research.

1. 01

   Manufacturing-aware generation

   Co-authored guided voxel diffusion research for porous microbial fuel-cell anodes: 64³ voxel designs with a learned minimum-feature-size surrogate guiding the later denoising stages.

2. 02

   3D perception and reconstruction

   Investigated single-view NeRF synthesis, diffusion-based reconstruction, quantized language models, and LoRA adaptation for 3D object generation and pose-estimation workflows.

3. 03

   Retrieval and agent systems

   Built database-connected agents using LangGraph, Tree-of-Thought, and ReAct; developed hybrid retrieval and HyDE-based RAG for business and process-optimization applications.

4. 04

   Research made explorable

   Developed an augmented-reality visualization of 186 neural-network results across 85 UCR time-series datasets for the World Map of AI research project.

Supervue AI

Lead Machine Learning Engineer

2020 — 2023

Lead ML Engineer · Feb 2021–Jan 2023 / ML Engineer · Jan 2020–Feb 2021

Led applied computer vision and language systems from model development through large-scale deployment.

1. 01

   Computer vision in production

   Led the AI application team and developed Supervue FR V2.0, a ViT-based facial-recognition API adopted by 50+ clients. Reduced API and application deployment costs by 10%. Built real-time zero-shot person detection and tracking with a CNN–Transformer encoder–decoder and TensorRT/ONNX inference, tracking 1.3M people with 99.1% uptime.

2. 02

   Systems and documents at scale

   Designed auto-scaling microservices for facial recognition, ASR, and social-distance monitoring, scaling AI applications to 30K users. Developed a distributed OCR network that improved document-parsing performance by 50% while processing 100M+ documents.

3. 03

   Task-specific language understanding

   Fine-tuned GPT-2 with Megatron-LM and DeepSpeed to build a multi-task chatbot NLU module, improving the intent-classification score by 10% over prior Bi-LSTM and GRU models.

4. 04

   Applied invention

   Developed a social-distance monitoring application using a patented one-shot object-flow regression method, granted as Indian patent 601573 (application 202041034063).

Samsung R&D

Research Collaborator

2021 — 2022

October 2021 – August 2022 · PRISM collaboration with Andhra University

Real-time video inpainting designed for the constraints of mobile devices.

1. 01

   Real-time visual reconstruction

   Engineered a multi-threaded inpainting algorithm using lightweight feature transformation, search modules, and caching; the profile reports 31 FPS and a 15% reduction in distortion artifacts.

2. 02

   From method to invention

   Contributed the nN-TFT approach for edge deployment, associated with the video-inpainting patent application reference 202241043300.

3. 03

   Industry–academic collaboration

   Conducted the research through Samsung R&D Institute Bangalore’s PRISM collaboration with Andhra University.

Zebi

Machine Learning Engineer

2018 — 2019

December 2018 – December 2019

Applied machine learning to identity, customer operations, and infrastructure reliability.

1. 01

   Faster identity workflows

   Helped build a blockchain-assisted check-in system with biometric verification and ID-card OCR APIs, reporting a 40% reduction in hotel check-in time.

2. 02

   Operational machine learning

   Built ticket summarization, build-failure prediction, and time-series alerting; the profile reports 0.95 recall for build-failure prediction and a 25% downtime reduction across 295+ servers.