Venkatesh Sankarappan

## Latent Ops - AI platform for knowledge work. # Building a change propagation engine for document heavy industries like MedTech. - Building a change propagation engine that navigates hierarchical, complex document structures, enabling LLMs to make precise auditable edits across large file sets. - Built an AI platform combining LLMs with a knowledge graph to help teams get source-grounded information from scattered workplace documents and tools. - Developed an agentic RAG system combining keyword and semantic search with a knowledge graph traversal to link internal data, documents, and workplace tools. - Built a multi agent orchestration system with custom agent runtime, handling context bloat, skills & tool registration, permissions, and tracing reasoning. - Built evaluation frameworks with binary pass/fail eval suites, step level tracing of agent execution paths, and evaluator skills for automated quality gates. - Designed data pipelines that process unstructured documents through LLM-based entity extraction, semantic chunking, and incremental sync from external APIs. For more detailed information, visit: https://latentops.ai/

Institute for Industrial Automation Systems - Federal Institute of Occupational Safety and Health (Buau): 1. Designed and implemented a synthetic data generation pipeline simulating human-robot interactions from 3D scenes in industrial environments to enhance the resilience of safety-critical vision systems. 2. Enhanced the robustness of AI models used for detecting workplace hazards, including PPE compliance and unsafe behaviors, directly supporting industrial safety monitoring. 3. Collaborated with cross-functional teams to create HSE-compliant vision datasets, advancing the development of human-centric operational workflows in automated manufacturing.

Topic: An Integrated Framework for Active Learning-Driven Data Sampling and Neural Architecture Optimization in Object Detection 1. Designed and implemented a production-grade deep learning-based data sampling framework for dense object detection, significantly improving MLOps efficiency and minimizing manual labeling requirements. 2. Built a scalable ML pipeline for iterative model refinement, utilizing curated datasets across training, deployment, and automated evaluation stages to drive continuous model performance improvements and accelerate feature iteration cycles. 3. Achieved comparable object detection performance using only 40% of the total training data, reducing development timelines by an estimated 30% and lowering labeling costs. Applications: faster model iteration cycles, context-aware feature development, efficient pre-labeling model workflows. Grade: 1.3 (max: 1.0)

Video-based Driver Assistance Systems - Deep Learning based Light Object Detection - Front Video Perception 1. Adapted and enhanced an internal data annotation tool to support the labeling process for new feature integration in video-based perception systems for automated driving. 2. Streamlined the ML workflow by improving annotation efficiency and accuracy, enabling faster and more reliable data preparation for training and validation pipelines.

Electric Power Train Solutions; Energy Management Product - Charcon & DC-DC converters 1. Developed a Human-Machine Interface (HMI) and automated test scripts for controlling and simulating system interactions in electric chargers and DC-DC converter applications for electric vehicles. 2. Streamlined manual testing procedures, achieving a 20% reduction in manual efforts and improving testing efficiency by approximately 15% per cycle, accelerating test case execution and overall validation timelines.

SynthiCAD: Synthetic Data Generation tool for Autonomous Industrial Robots performing multiple computer vision tasks 1. Developed a Python-based tool to generate realistic, labeled synthetic dataset from 3D object models, optimized for integration with task-specific neural networks. 2. The tool’s significant impact was documented in a research paper and contributed to an open-source dataset, addressing rare industrial use cases in computer vision. Grade: 1.0 (Max: 1.0)

Job Description: * Functional Software Developer for Electric Power Steering Systems(ASIL-D), predominantly worked in VW MQB,MEB platform projects * Collaborated for Customer Requirements Analysis and Technical realization of concept * Worked in RDCT phases of Software Development LifeCycle * Good knowledge of ASPICE & ISO26262 Safety Standards * Hands-on experience in CAN & CANFD communication protocols * Predominantly worked on Advanced Driver Assistance Systems (Lane Keeping Assist, Lane Departure Warning, Parking Assist, Driving Steering Recommendation) Roles & Responsibilities: + Functional development using MATLAB,Simulink,TargetLink (Auto-Code Generation) + Performed MIL,SIL and BTC Embedded Testing + Modeling Guidelines(MAAB,MISRA) and ISO26262 Safety Standards + Performed Static Code Analysis using ASTREE and QA-C + Carried out Functionality Test and Integration Testing on the bench using vector tools + Problem Job Analysis and supported system level testing (HIL) Tools Used: - Design Tools ~ MATLAB (Simulink, Stateflow) ~ Enterprise Architect - dSpace ~ TargetLink, BTC Embedded Tester - Vector Tools ~ CANoe, CANalyzer,and CANape - Debugging Tool ~ Lauterbach(Trace32) - IBM Tools ~ ClearCase, ClearQuest and DOORS - Unit-testing Tools ~ ASTREE, VCar

- Hands on experience with PLC controllers and Distributed Control Systems - Completed mini project on control of servo motor using PLC

Specialization: Machine Learning, Deep Learning, Computer Vision

Bachelor Thesis: Smart Surveillance System