★ 2× ACCEPTED PAPERS — ICSoftComp 2026 (Springer CCIS): AdaptiLLM & NC-MedAI ★ BEST PAPER AWARD — ComSIA 2026 (Springer) ★ ACCEPTED — NC-MedAI Deep Learning @ ICSoftComp 2026 (Springer CCIS) ★ ACCEPTED — AdaptiLLM @ ICSoftComp 2026 (Springer CCIS) ★ WINNER — State-Level GenAI Hackathon 2026 ★ WINNER — State-Level Agentic AI Hackathon 2026 ★ GLOBAL RANK 15 — WiDS Datathon Challenge 2025 ★ AWS STUDENT BUILDER GROUPS CHAPTER LEAD ★ 2× ACCEPTED PAPERS — ICSoftComp 2026 (Springer CCIS): AdaptiLLM & NC-MedAI ★ BEST PAPER AWARD — ComSIA 2026 (Springer) ★ ACCEPTED — NC-MedAI Deep Learning @ ICSoftComp 2026 (Springer CCIS) ★ ACCEPTED — AdaptiLLM @ ICSoftComp 2026 (Springer CCIS) ★ WINNER — State-Level GenAI Hackathon 2026 ★ WINNER — State-Level Agentic AI Hackathon 2026 ★ GLOBAL RANK 15 — WiDS Datathon Challenge 2025
COMPUTER SCIENCE & AI @ KLE TECH UNIVERSITY

Architecting
Distributed Cloud Systems
& Intelligent AI Engines.

Backend Engineer, Cloud Architect, and AI Systems Researcher. Author of two research papers accepted at ICSoftComp 2026 (Springer CCIS) covering adaptive on-device LLM inference (AdaptiLLM) and geometric deep learning for medical vision (NC-MedAI), and recipient of the ComSIA 2026 Best Paper Award.

VIEW PROVEN SYSTEMS → GITHUB REPOS → LINKEDIN PROFILE →
2× ACCEPTED PAPERS @ ICSOFTCOMP 2026 (SPRINGER CCIS)
1st AUTHOR COMSIA 2026 BEST PAPER AWARD (SPRINGER)
2× STATE HACKATHON WINNER (2026)
RANK 15 WIDS GLOBAL DATATHON 2025
CHAPTER 01

Proven Systems & Infrastructure.

End-to-end architectures designed for data pipeline orchestration, on-device LLM inference, medical deep learning, GPU computing, and reinforcement learning.

01
FASTAPI SPARK MINIO POSTGRESQL DOCKER

DataForge — Enterprise Data Pipeline Orchestrator

Data pipeline orchestration platform supporting dataset ingestion, Spark-based ETL processing, asynchronous job execution, and metadata management.

Key Architecture & Impact:
  • Designed reusable Spark transformation workflows for cleaning, validation, aggregation, and schema enforcement of uploaded datasets.
  • Integrated MinIO for object storage and PostgreSQL for metadata and job tracking.
  • Implemented Docker-based service orchestration for reproducible deployment across local and cloud environments.
GITHUB REPO ↗
02
ACCEPTED @ ICSOFTCOMP 2026 (SPRINGER CCIS) ANDROID KOTLIN C++ JNI TFLITE QWEN2.5 QUANTIZATION

AdaptiLLM — Adaptive On-Device LLM Inference

Adaptive on-device LLM inference system that dynamically selects inference configurations based on query complexity and device resource constraints.

Key Architecture & Impact:
  • Springer CCIS Accepted: Research paper on AdaptiLLM accepted for publication at ICSoftComp 2026 (Springer CCIS).
  • Implemented a policy engine that adapts CPU threads, context length and generation limits using latency, energy, and quality signals.
  • Built a lightweight TFLite query classifier for workload characterization and real-time resource allocation.
  • Evaluated multiple inference modes on a quantized Qwen2.5 model, achieving improved latency and energy efficiency across 720 inference runs.
GITHUB REPO ↗
03
ACCEPTED @ ICSOFTCOMP 2026 (SPRINGER CCIS) PYTORCH DEEP LEARNING NEURAL COLLAPSE ETF HEADS MEDICAL AI

NC-MedAI — Neural Collapse Medical Deep Learning

Deep learning research framework investigating penultimate feature geometry degradation in severely class-imbalanced medical image datasets (HAM10000).

Key Architecture & Impact:
  • Springer CCIS Accepted: Research paper accepted for publication at ICSoftComp 2026 (Springer CCIS).
  • Formulated real-time feature geometry tracking across within-class scatter (NC1), ETF alignment (NC2), and classifier weight-feature dual alignment (NC3).
  • Engineered fixed Equiangular Tight Frame (ETF) heads that mathematically enforce maximum pairwise class separation.
  • Eliminated majority-class gradient domination to slash diagnostic false-negative rates on critical conditions (Melanoma).
GITHUB REPO ↗
04
PYTHON FASTAPI REACT GRAPHRAG FAISS BM25 NETWORKX JWT

Synapse AI Tutor — Adaptive AI Tutoring Platform

Adaptive AI tutoring platform that identifies learner knowledge gaps and adapts explanations to proficiency levels using LLMs and GraphRAG-based retrieval.

Key Architecture & Impact:
  • Implemented hybrid retrieval using FAISS vector search, BM25 lexical search, and NetworkX knowledge graphs to provide context for LLM responses.
  • Engineered automated learner gap identification algorithms that tailor explanations dynamically based on real-time feedback loops.
  • Built secure REST endpoints with FastAPI and JWT authentication for isolated student session management.
ADDITIONAL HIGH-IMPACT SYSTEMS & GITHUB CODEBASES ALL 50+ REPOS ON GITHUB →
05
CUDA 12.X C++17 GPU OPTIMIZATION

CUDA Parallel Reduction Suite

7-level progressive CUDA GPU optimization study analyzing kernel execution timing, PCIe H2D/D2H transfer latency, warp divergence elimination, shared memory bank conflict reduction, and RTX 3050 peak memory bandwidth utilization.

GITHUB ↗
06
BEST PAPER @ COMSIA PYTORCH GYMNASIUM

Dynamic Pricing with Deep RL

Research codebase backing the ComSIA 2026 Best Paper Award. Simulates Poisson customer arrivals, Multinomial Logit price elasticity, stockout constraints, and multi-agent RL (PPO, SAC, DDPG, LinUCB).

GITHUB ↗
07
YOLOV8 OPENCV PYTHON

Watchman — Edge Vision CCTV

Zero-hardware-cost AI surveillance platform converting Android phone hardware into an intelligent CCTV feed with real-time YOLOv8 human detection, low-latency threaded MJPEG capture, and native desktop alerts.

GITHUB ↗
08
ANDROID TFLITE MOBILENET V2

Mobile Object Detection (TFLite)

Real-time on-device computer vision application deploying MobileNet v2 via TensorFlow Lite to classify live Camera2 preview frames into 1,000 ImageNet classes with 500ms intervals entirely offline.

GITHUB ↗
09
FASTAPI REACT REDIS

DocChat — Full-Stack RAG Q&A

Asynchronous document intelligence platform supporting multi-format document ingestion, Redis vector embedding caching, PostgreSQL metadata management, and JWT-authenticated tenant workspaces.

10
KUBERNETES OPEN5GS PROMETHEUS

5G Network Slice Orchestrator

Kubernetes 5G slice testbed simulating eMBB/URLLC/mMTC traffic using Open5GS and UERANSIM with automated real-time Prometheus QoS monitoring and dynamic pod autoscaling.

CHAPTER 02

Research Output & Major Honors.

Academic publications in Springer journals & CCIS series, top-tier global hackathon wins, and competitive ML datathon titles.

SPRINGER CCIS PUBLICATION

Accepted — ICSoftComp 2026 (AdaptiLLM)

Author · Springer CCIS Series

Research paper titled "AdaptiLLM: Adaptive On-Device LLM Inference" accepted for publication in Springer CCIS (Communications in Computer and Information Science) series. Evaluated adaptive multi-objective configuration on quantized Qwen2.5 on-device across 720 benchmarked runs.

SPRINGER CCIS PUBLICATION

Accepted — ICSoftComp 2026 (Deep Learning / NC-MedAI)

Author · Springer CCIS Series

Research paper titled "NC-MedAI: Neural Collapse-Aware Medical Image Classification" accepted for publication in Springer CCIS series. Diagnoses penultimate feature geometry degradation under extreme medical dataset imbalance (HAM10000) using Neural Collapse metrics and enforces Equiangular Tight Frame (ETF) classifier heads.

SPRINGER PUBLICATION · BEST PAPER AWARD

Best Paper Award — ComSIA 2026

First Author · Springer Conference Series

Awarded Best Paper Award for original research paper titled "Reinforcement Learning for Dynamic Pricing". Developed multi-agent RL algorithms optimizing revenue & demand elasticity under stochastic constraints.

GLOBAL DATATHON

Global Public Rank 15 — WiDS Datathon 2025

Women in Data Science (WiDS) Challenge

Achieved Global Public Rank 15 out of thousands of international teams by building robust gradient-boosted ensembles and rigorous cross-validation feature pipelines.

STATE HACKATHON WINNER

Winner — GenAI Hackathon 2026

State-Level Competition · KLE Tech

Secured 1st Place overall for engineering a real-time Generative AI workflow application under tight sprint constraints.

STATE HACKATHON WINNER

Winner — Agentic AI Hackathon 2026

State-Level Competition · KLE Tech

Secured 1st Place overall by building multi-agent collaborative tools featuring dynamic task delegation and local model invocation.

CHAPTER 03

Technical Mastery & Tooling.

Proven stack across system design, backend engines, cloud platforms, and data infrastructure.

01

LANGUAGES & CORE CS

Java Python CUDA C++ SQL HTML/CSS C++ Kotlin DSA OOP Operating Systems Computer Networks
02

BACKEND & DISTRIBUTED

FastAPI REST APIs Microservices PostgreSQL Redis JWT Authentication
03

DATA ENGINEERING & STORAGE

Pandas Apache Spark ETL Pipelines MinIO Object Storage
04

FRONTEND, DEVOPS & ML/AI

React Docker Git / GitHub CI/CD PyTorch CUDA GPU Computing LLM Inference llama.cpp GraphRAG FAISS BM25 NetworkX TFLite Qwen2.5 Quantization Neural Collapse Gymnasium / RL YOLOv8 OpenCV
CHAPTER 04

Leadership & Academic Credentials.

Guiding developer communities, leading AWS chapters, and academic rigor at KLE Technological University.

LEADERSHIP & COMMUNITY

AWS Student Builder Groups — Chapter Lead KLE Technological University

Directed campus AWS initiatives. Led hands-on cloud architecture workshops covering AWS EC2, S3, AWS Lambda, and serverless design patterns, mentoring 50+ students in production deployments.

Google Student Ambassador KLE Technological University

Represented Google developer platforms and technology programs across campus, spearheading student technical bootcamps and developer activities.

ACADEMIC FOUNDATION

KLE Technological University

2023 — 2027
B.E. Computer Science (Artificial Intelligence)
Core Curriculum: Data Structures & Algorithms, Object-Oriented Programming, Operating Systems, Computer Networks, Database Management Systems.
CHAPTER 05

Initiate Dispatch.

Available for high-impact backend engineering roles, cloud architecture projects, and AI system collaborations.

EMAIL DISPATCH →
rishipkulkarni@gmail.com
DIRECT PHONE →
+91 88616 06007
LINKEDIN NETWORK →
linkedin.com/in/rishipkulkarni
GITHUB CODEBASE →
github.com/geeky-rish