Konark Verma

SDE-3, Team Lead, Denso Japan

Masters in CSE, IIT-Delhi

I build high-performance systems at the intersection of AI, cloud infrastructure, and IoT.

About Me

I'm a software engineer with passion for building edge-native, end-to-end AI solutions using Small Language Models (SLMs), Parameter Efficient Finetuning Techniques (PEFT) like LoRA/QLoRA, advanced RAG pipelines and Agentic AI workflows.

With 5+ years of experience, I also specialize in backend architecture, AWS services (using CDK), and CI/CD-driven deployments for scalable, reliable systems.

I hold an Masters Degree in Computer Science from IIT Delhi and focus on building scalable, production-grade AI systems from concept to deployment.

Here are a few technologies I've been working with:

  • ▹Agentic AI
  • ▹Small Language Models (SLM)
  • ▹PEFT (LoRA/QLoRA)
  • ▹LLM Orchestration
  • ▹Vector Databases
  • ▹RAG
  • ▹Microservices Architecture
  • ▹Docker
  • ▹AWS (ECS, ECR, S3, Bedrock)
  • ▹AWS CDK
  • ▹CI/CD Pipelines
  • ▹Edge AI

Experience

  1. DENSO Corporation JapanTokyo, Japan

    Software Development Engineer-3April 2025 – Present

    Product Owner, Team Lead

    Designed and built an on-device automotive AI assistant platform leveraging advanced RAG pipelines with query routing and parameter-efficient fine-tuning (LoRA/QLoRA) on Small Language Models to deliver low-latency, vehicle-specific intelligence.
    Engineered a real-time voice AI orchestration system integrating STT, TTS, wake-word detection, VAD, and LLM-based tool execution, optimized for edge deployment (e.g., Raspberry Pi) with robust acoustic processing and asynchronous control flows.
    Software Development Engineer-2April 2022 – March 2025

    Product Owner, Team Lead

    Proposed the idea and led development of Quad-Installer (NPM CLI for automated edge deployment) and Quad-Terminal (web-based terminal for interacting with Dockerized microservices on edge devices).
    Improved system reliability, automated lifecycle management, and supported integration of Docker, AWS, and modern DevOps practices in automotive and IoT projects.

  2. DIIN, Denso International IndiaGurugram, Delhi-NCR, India

    Software Development Engineer-1October 2020 – March 2022
    Built an NLP-based Similar Companies Recommendation System using GloVe Embeddings, TF-IDF, Hierarchical Clustering. Developed and deployed a Django based web app via Docker, AWS ECR and ECS to enable company recommendation for business discovery.
    Implemented autonomous driving steering system using Conv2D, Conv3D, Conv-LSTM models, achieving up to 98% accuracy from live video feeds.

  3. Tensor Dynamics Private LimitedIIT Delhi, Delhi, India

    Project InternOctober 2018 – December 2018
    Built statistical visualization models and deep learning systems to forecast solar power day ahead using ARIMA and LSTM sequence models on historical solar data.

  4. Plaksha Technology Leaders FellowshipGurugram, Delhi-NCR, India

    Teaching AssistantSeptember 2020 – March 2021
    Helped design and teach courses in Machine Learning and Deep Learning.
    Created assignments, quizzes, and exams for students.
View Full Résumé

Projects

  1. AutoEdge AI: On-Device Automotive Assistant Platform

    AutoEdge AI empowers OEMs to seamlessly transform vehicle manuals and diagnostic data into deployable, vehicle-specific AI assistants.
    Leveraging a hybrid architecture - a base automotive Small Language Model (SLM) combined with LoRA adapters and on-device Retrieval-Augmented Generation (RAG) - our platform enables OEMs to 'walk away' with a personalized and optimized AI solution that runs on edge.

    • Python
    • PEFT
    • QLoRA
    • RAG
    • Small Language Models
    • Vector DB
    • Chroma
    • PyTorch
    • Edge AI
    • Document Parsing
    • Embedding Models

  2. Quad-AI Voice Assistant System

    A real-time voice-based conversational AI orchestrator for Denso's Quad (formerly Parker) Vehicle System, that runs on Raspberry Pi.
    Supports wake word detection, VAD, STT (AWS Transcribe, Faster-Whisper), TTS (AWS Polly), LLM integration via AWS Bedrock, AEC, and MCP tools for synchronous/asynchronous execution with progress polling.

    • Python
    • Rust
    • LLM
    • AWS Bedrock
    • STT
    • Faster Whisper
    • AWS Transcribe
    • TTS
    • AWS Polly
    • Wakeword Detection
    • PV-Porcupine
    • AEC
    • VAD
    • MCP
    • Docker
    • Raspberry Pi
    • OpenAI API

  3. Quad-Terminal

    A web-based terminal for Denso's Quad (formerly Parker) Vehicle System, that runs on edge devices (Raspberry Pi, QualComm, Nvidia Jetson).
    It enables interaction with Dockerized microservices via WebSockets and remote cloud access through port forwarding for debugging and testing purposes.

    • React
    • TypeScript
    • NodeJS
    • Docker
    • AWS ECS
    • AWS ECR
    • WebSockets
    • gRPC
    • AWS CDK
    • REST APIs
    • NGINX
    • Jest

  4. Quad-Installer

    An NPM-based CLI tool for Denso's Quad (formerly Parker) Vehicle System, that handles lifecycle management of Dockerized microservices on edge devices.
    Enables remote monitoring, automated deployment, and self-updating edge infrastructure via seamless cloud integration - managing Docker containers and systemd services with built-in job scheduling, prioritization, and continuous optimization.

    • TypeScript
    • NodeJS
    • Docker
    • SystemCTL
    • AWS S3
    • AWS Cognito
    • AWS CDK
    • gRPC
    • Jest

  5. District Development Model for India

    Used Mass Media and Census data to develop a method for identifying relevant news articles that explain and analyze growth in Indian districts across Agriculture, Development, Lifestyle, Environment, and Industrialization.

    Published in DSSG-21, KDD 2021. Provided actionable insights for district-level development analysis.

    • Doc2Vec
    • DocTag2Vec
    • TF-IDF
    • Hierarchical Clustering
    • K-Means Clustering
    • Glove-Embeddings

  6. Similar Companies Recommendation System

    Enabled automated discovery of similar companies for business intelligence and market research.
    NLP-based system that recommends similar companies using Glove Embeddings, TF-IDF, Hierarchical Clustering, K-Means Clustering, and Cosine similarity.
    Includes a Django web app deployed with Docker, AWS ECR, and AWS ECS.

    • Python
    • NLP
    • Glove Embeddings
    • TF-IDF
    • Hierarchical Clustering
    • K-Means Clustering
    • Cosine Similarity
    • Django
    • Docker
    • AWS

  7. Latex to HTML Compiler/Converter

    Compiler to convert Latex code to HTML code.
    Implements Abstract Syntax Tree along with the grammar.
    Uses Flex and Bison as Lexical and Syntax Analyzer.

    • C
    • Flex
    • Bison
    • Compiler Design
    • Abstract Syntax Tree
View Full Project Archive

Education

  1. 2018 - 2020

    M.Tech. — Computer Science and Engineering

    Indian Institute of Technology, Delhi

  2. 2013 - 2017

    B.Tech. — Computer Science

    University of Delhi

Achievements

  3. Achievement

    Achieved AIR (All India Rank) 453 (99.6 percentile) in GATE 2018 among 107,893 candidates.

  5. Achievement

    Won Game Development Hackathon by redesigning 'Paratroopers' with multiple levels and power-ups using Unity3D.

  6. Achievement

    Secured 3rd position at Metlife Japan Hackathon, by designing and deploying a policy management website for client and agents.

Contact Me

Always open to discussing new projects, creative ideas, or opportunities.

Reach me via LinkedIn, GitHub or directly mail me at kon172verma@gmail.com.