Profile
Electrical + CS Engineering undergraduate (Rank 1, CGPA 9.39/10) specializing in Large Language Models, Multimodal AI, and Edge-Optimized Inference. Experienced in fine-tuning and deploying large-scale LLMs on accelerators, integrating ML into high-performance systems, and developing award-winning AI applications for social good.
Passion Areas: cutting-edge AI/ML, particularly Large Language Models, adaptable high-performance scalable inference, brain computer interfaces and human-centered AI.
Education
Dayalbagh Educational Institute, Agra 2023 – 2027
Bachelor of Technology — Major: Electrical Engineering · Minor: Computer Science
Rank: 1/328 (Engineering) · CGPA: 9.391/10
Co-Curriculars: Social Service (NSS), Debate Team, Sitar (Indian Classical), Local sustainable farming community
Co-Curriculars: Social Service (NSS), Debate Team, Sitar (Indian Classical), Local sustainable farming community
Apeejay School, Noida 2009 – 2023
High School Diploma
President – Code Club 2023 · President – Robotics Club 2022 · Head Boy – Student Council 2018, 2015
Work Experience
LLM Infra Intern — Gatespeed, Pleasanton CA Aug – Oct 2025
Intel AI Partner Alliance Member · Multilingual RAG & High-Performance LLM Deployment
- Built multilingual RAG with semantic retrieval using Qdrant.
- Deployed optimized inference of DeepSeek R1 Llama 70B on Intel Gaudi2 cluster (8×98GB HPUs) with hybrid RTX (2x5070ti + 5090ti) acceleration via vLLM/Ollama.
- Fine-tuned DeepSeek-R1-Distill-Llama-70B on Gaudi2, achieving 73.77% accuracy, 2.72 perplexity in 52min with 16.4M trainable parameters; optimized GPT-2 to 26.95 samples/sec.
- Enhanced inference with HPUgraph, KV caching, containerized scaling, and ZeRO Stage 3, sustaining 94.1GB peak memory per HPU (bfloat16).
- Cut e2e latency from 261s → 46s; reached 1199 tokens/s output, 6691 tokens/s total throughput at 256 req/s concurrency.
- Stack: Intel Gaudi2 HPU, vLLM, Ollama, Qdrant DB, DeepSeek LLMs, LoRA/PEFT, GGUF quantization, Docker, Habana Frameworks, DeepSpeed ZeRO, multilingual NLP.
Machine Learning Trainee — Cadence Design Systems, Noida Jun – Aug 2025
EDA Tool Optimization via ML — Fortune 500 Semiconductor Company
- Applied ML-based predictive filtering to speed up ECO optimizer runtime on Samsung, Qualcomm and Renesas 3nm chips with no QoR degradation. Filtered 65% wasteful evaluations on avg.
- Developed and integrated feature engineering and training pipeline (Random Forest, LightGBM, XGBoost) into Cadence C/C++ code for Tempus and Certus signoff tools.
- Automated log parsing, profiling, and data preparation using Bash, CSH and Tcl/Tk.
- Stack: Python, scikit-learn, LightGBM, XGBoost, C/C++, Tcl/Tk, Shell, NumPy, Pandas, Cadence EDA tools.
Intern — MindLab, IIT Delhi May – Jun 2025
LLM & Multimodal AI for Neuroscience & Therapeutic Gaming
- Prototyped adaptive NPC dialogue systems using fine-tuned LLMs (DeepSeek-V3-70B, Mistral-8B) with OCEAN personality modeling and vector-based memory.
- Implemented weighted temporal graphs for multi-agent knowledge diffusion with personality-filtered propagation and distributed vector databases.
- Built FastAPI middleware around OpenRouter with vanilla JS/HTML test UI, integrating GQ-6 metrics for therapeutic RL optimization targeting gratitude learning.
- Stack: FastAPI, OpenRouter API, Ollama, DeepSeek-V3-70B, Mistral-8B, Python, Qdrant, temporal graph algorithms.
Intern (Systems & Networking) — Omnitech Solutions, San Jose CA May – Jul 2024
High-Performance Networking, Linux/BSD, Virtualization
- Setup NGINX media streaming server on Intel Xeon Gold CPUs with high-speed network cards (e810-cqda2, xxv710) running low-latency media streaming protocols.
- Deployed Linux VMs with KVM/QEMU, virtual networks; utilized DPDK and custom kernel device drivers for high-speed packet processing.
Intern (DevOps & Backend) — QuditBrain, Noida Jun – Aug 2023
Backend APIs, Server-Side Systems
- Managed AWS infrastructure (EC2, S3, Lambda, DynamoDB, SQS, Route 53, Cognito, CloudWatch); automated via Boto3 and AWS APIs.
- Built and deployed Python backend services, secured access-control workflows, integrated WireGuard VPNs with AWS IAM-aligned policies.
- Developed monitoring and logging pipelines; improved backend reliability across distributed systems.
Projects & Research
deGuppe TOR · Blockchain · Python · WebSockets · SQLite
- Decentralized, peer-run, real-time communication system over TOR with hybrid blockchain storage.
- Best Poster Presentation — 47th National Systems Conference, Systems Society of India.
- International Soonami Cohort 3 funding · Best Project Web3/AI for Good · Third Prize IITD Tryst Track, Best Live Demo.
Gam-i-yog Python · OpenCV · MediaPipe · PyTorch · Multi-modal GenAI
- Live pose classification and multimodal feedback using GenAI; ensemble with dynamically trainable pose classifier, K-means clustering, and generative personalized feedback.
- Best Poster Award — DSC Conference (University of Waterloo, CAU Kiel, Western University, University of Birmingham).
- Grand finale winners of National Toycathon Hackathon; received Government of India funding for impact on Yoga practitioners including accessibility for injuries and disabilities.
ESG–Financial Performance Research Python · Bloomberg Terminal · Statistical Analysis
- Empirical analysis of ESG score correlation with firm financial performance; imputation algorithms for missing ESG data.
- Published in Springer Cureus journal via ICASSSD Summer School 2024 collaboration.
Abhinandan (War Robot) Mechatronics · CAD · Combat Robotics
- First Prize — IITD National Tryst RoboWars. Non-violent defensive battlebot with frugal attack-resilient design.
- First junior high school team to beat funded collegiate teams nationally.
Pehchaan PyTorch · EdgeFace · OpenCV · Raspberry Pi · Embedded Linux
- Edge-optimized face recognition using EdgeFace model (LoRaLin distilled layers); deployed on Raspberry Pi for security turnstiles and classroom attendance.
Technical Skills
Programming & Dev Python, C/C++, Rust, Flask, FastAPI, SQL
Systems & Networking Linux/BSD, Shell, NGINX, Docker/Kubernetes, Grafana/Prometheus, QEMU/KVM, WireGuard, AWS (EC2, Lambda, S3, DynamoDB, RDS, Aurora, Route 53, Cognito), Wireshark, Nmap, OpenSSL
ML & Computer Vision TensorFlow, PyTorch, OpenCV, MediaPipe, EdgeFace, K-Means, Ollama, vLLM, Unsloth, multimodal LLMs, world models, vision-language-action models
Blockchain & Web3 Hybrid Blockchain, Proof of History, decentralized systems design, TOR, TON blockchain, Solana