SV
SRIKAR
VELUVALI
IVth Year @ KMIT
About Me
About - Srikar Veluvali
×
👋 Hello, World!
I'm Srikar Veluvali, currently pursuing a Bachelor of Technology in Information Technology at Keshav Memorial Institute of Technology with a CGPA of 9.53.
I'm currently an SDE Intern at Amazon, where I built dispute & claims management software serving 3.7M+ auditors. Before that, I was a Research Intern at Microsoft in Bengaluru, and earlier an HPC Software Engineering Intern at DRDL (DRDO).
I've built projects like Astor AI, a medical chatbot fine-tuned on Llama 3 with Retrieval-Augmented Generation achieving 800+ downloads, and a Heart Health Web App using the MERN stack and machine learning with Google's Gemini AI for personalized plans, nominated for the People's Choice Award in the Google Gemini API Developer Competition.
When I'm not coding or tinkering with tech, I enjoy solving problems on LeetCode (1,172+ problems, rating 1713) or exploring the latest in generative AI.
Academic Performance
Overall CGPA
9.53
Branch Topper - All Semesters
Year 1
Sem 1:9.84
Sem 2:9.79
Year CGPA:9.81
Year 2
Sem 1:9.62
Sem 2:9.86
Year CGPA:9.74
Year 3
Sem 1:9.26
Sem 2:9.37
Year CGPA:9.31
Year 4
Sem 1:9.05
Sem 2:9.47
Year CGPA:9.26
Skills & Expertise
Work Experience
SDE Intern
Amazon | Jan 2026 - Jun 2026
- •Engineered auditing dispute & claims management software serving 1200+ auditors across Amazon, streamlining resolution workflows and reducing auditing time through automated processing pipelines
- •Owned the end-to-end client onboarding lifecycle from intake to activation, collaborating cross-functionally with engineering and operations teams to standardize workflows and accelerate time-to-productivity for new accounts
- •Migrated user access management from individual Login IDs to LDAP and POSIX group-based authorization, enabling centralized permission control and improving scalability, consistency, and governance across Controllership teams
- •Eliminated the 7-day onboarding SLA entirely by designing and implementing group-based access provisioning, empowering teams to self-manage memberships and significantly reducing IT dependency for access changes
Research Intern
Microsoft | Jan 2025 - Jul 2025
- •Designed a graph-based code optimizer that auto-transforms NVIDIA's PTX assembly for faster execution
- •Built a modular pipeline featuring parsing, SSA conversion, control-flow graph generation, and symbolic analysis
- •Developed advanced pattern-detection algorithms via symbolic execution to spot non-coalesced memory accesses, then automated kernel rewrites to enforce optimal memory-access patterns and boost throughput
- •Worked with engineers and researchers to handle PTX syntax variations across multiple architectures
- •Achieved a 74.12% performance improvement in matrix multiplication kernels through these transforms
HPC Software Engineering Intern
Defence Research Development Laboratory (DRDO) | Jun 2024 - Dec 2024
- •Cut GPU runtime for missile-flow CFD simulations by 25.93%—from 30 days down to 22
- •Crafted and tuned parallel algorithms to drive higher GPU-core utilization
- •Partnered closely with engineers to align compute kernels with fluid-dynamics requirements
- •Accelerated overall project timelines through leaner, more efficient simulation workflows
Featured Projects
$cat README.md
Built a GRPO reinforcement-learning pipeline on Qwen2.5-3B (GSM8K) with multi-objective rewards—XML schema, integer check, and exact-match—to enforce structured chain-of-thought reasoning. Fine-tuned with 4-bit QLoRA/LoRA (Unsloth/PEFT), vLLM on-policy sampling, and TRL's GRPOTrainer on a single GPU, strengthening a 3B model's math reasoning by 4x using a DeepSeek-inspired, schema-aligned methodology.
$ls tech-stack/
PythonPyTorchQwen2.5Reinforcement LearningGRPOQLoRA / LoRAUnslothPEFTvLLMTRL
$git remote -v
$
$cat README.md
Built a YAML-driven async graph workflow engine using Kahn's topological sort and asyncio.gather to auto-parallelize DAG stages—supporting conditional branching, foreach/range/while loops, retries, and timeouts across 22 node types (HTTP, LLM, file I/O, data transforms). Engineered a multi-agent self-healing subsystem (ReAct pattern) with a Monitor → Diagnostician → Repair pipeline that parses failures across 7 languages, generates LLM-guided patches with rollback safety, and re-verifies fixes autonomously. Ships with a unified LLM provider abstraction across 5 backends, git-backed versioning, cron scheduling, and 199+ pytest tests.
$ls tech-stack/
PythonFastAPIasyncioLLM OrchestrationReActOpenAIAnthropicGroqOllamaAWS Bedrock
$git remote -v
$
$cat README.md
Built PRISM, a comprehensive learning platform that leverages AI to transform how students study and prepare. Features include RAG-based document Q&A, AI-powered quiz generation, virtual interview practice, interactive flashcards, PDF annotation, mind maps, and progress analytics. Supports multiple formats including PDFs, Word documents, text files, and YouTube videos.
$ls tech-stack/
ReactFastAPIGroq APIPineconeMongoDBTipTapSentenceTransformersJWTGoogle OAuth
$git remote -v
$npm run deploy
$
$cat README.md
Developed Searchly, an AI-driven e-commerce assistant built on a robust microservices architecture with an agentic workflow that orchestrates data scraping, embedding, retrieval, and UI interactions. Features include "Aivy" natural-language chat, personalized recommendations via HuggingFace embeddings and Pinecone vector search, real-time product data scraping, JWT-based authentication, and a responsive React/Tailwind UI.
$ls tech-stack/
FlaskPythonMongoDBPineconeHuggingFace EmbeddingsGroq APIBeautifulSoupReactTailwind CSS
$git remote -v
$play demo.mp4
$
$cat README.md
Built a medical chatbot powered by the Llama 3 architecture with Retrieval-Augmented Generation (RAG) for accurate, evidence-backed clinical Q&A. Fine-tuned on curated medical datasets and published on Hugging Face, reaching 800+ downloads.
$ls tech-stack/
React.jsFlaskLLMsGenerative AI
$git remote -v
$play demo.mp4
$
$cat README.md
Developed a MERN stack web app with Machine Learning for heart disease prediction, offering personalized diet and exercise plans and integrating Google's Gemini AI for diet suggestions.
$ls tech-stack/
MERNMachine LearningGoogle Gemini AIFlask
$git remote -v
$play demo.mp4
$
$cat README.md
Developed an Optical Character Recognition (OCR) system using PaddleOCR combined with a fine-tuned BERT model to extract and classify entities from text in images.
$ls tech-stack/
Optical Character RecognitionBERTPythonMachine Learning
$git remote -v
$
$cat README.md
Developed a Pokemon-style RPG game using Java and Object-Oriented Programming concepts. Players engage in turn-based battles and progress through various levels with different strategies.
$ls tech-stack/
JavaOOPS
$git remote -v
$
$cat README.md
A Python-based data analysis project that explores the Video Game Sales Analysis dataset to answer 15 key questions related to video games. Presented at PRAKALP 2023.
$ls tech-stack/
PythonData AnalysisVisualizationTableau
$git remote -v
$
$cat README.md
AI-powered navigation assistants. LUNA is voice-based, while LEVI is text-based, designed to assist users with website navigation, answering queries, playing music, and more.
$ls tech-stack/
AIPython
$git remote -v
$
$cat README.md
A text-based command-line game in C, simulating battles between the player and various bots. The game features strategic combat and a secret battle with a special challenge.
$ls tech-stack/
CCommand-line
$git remote -v
$
Education
2020
Class 10
Dilsukhnagar Public School
2022
Class 12
Sri Chaitanya Junior College
Expected 2026
B.Tech - IT
Keshav Memorial Institute of Technology
Branch Topper
Achievements & Certifications
Consistently achieved the highest academic standing as the Branch Topper across all semesters.
Srikar Veluvali
1 / 9