Building bots that nearly pass the Turing test - Go panics, GPU jobs, and hotfixes included.
Backend mayhem or agents hallucinating? I’ve got caffeine, let’s cook.

Org I'm building: HyperKuvid-Labs

• Backend & Distributed Systems: Building robust APIs (FastAPI/Go), container orchestration, and microservices that don't crash under load.
• High-Performance Compute: Optimizing inference engines, managing GPU scheduling, and squeezing performance out of hardware.
• Fullstack Engineering: Connecting the dots from the database (Postgres/Mongo) to the client (React/TS), handling the chaos in between.

Python • Docker • Nosana • SGLang

• Backend for decentralized GPU compute
• GPU job scheduling and container orchestration
• 30% lower latency, 40% better GPU usage

TypeScript • Go • MCP Protocol

• Infra middleware for AI agents
• Programmatic job and resource control
• 40% faster jobs, 25% lower infra cost

FastAPI • C++ • React • Prisma • Docker

• High-performance stowage optimization system
• Python backend driving C++ compute
• 10M+ items processed in under 5 seconds

IEEE ICICIS 2025 (Accepted)

• Architected edge-native NLP routing on Raspberry Pi 5 using NER/syntactic analysis, slashing inference time by 21.34% with <2.7% relevance loss.

• Devised adaptive thresholding (EMA α=0.2) and cosine similarity logic to dynamically route requests across tiered models for zero-latency on-device execution.

IEEE ICPC2T 2026 (Accepted)

• Engineered adaptive speculative decoding with multi-parent quantization (FP16/Q8/Q4), achieving 22.6% faster inference via hardware-aware draft–target verification pipelines.
• Implemented complexity-based prompt classification (syntactic/semantic metrics) for dynamic precision routing, maintaining <2% accuracy loss across diverse benchmarks.

• Inference Engines: vLLM, SGLang, Ollama, ONNX
• Optimization Techniques: EAGLE, Speculative Decoding, Quantization, LoRA/PEFT
• Compute & Systems: CUDA, C/C++, Distributed Training
• Vision & Architectures: PyTorch, OpenCV, YOLO, CNN, ViT
• Applied AI: Reinforcement Learning (RL), Diffusion Models, Speech and Summarization Systems
• AI Orchestration: LangChain, LangGraph, Model Context Protocol (MCP), CrewAI

• Core Stack: Python (FastAPI/Django), Go, C++
• Data Layer: PostgreSQL, MongoDB, Redis, Prisma
• Frontend: React.js, Tailwind CSS, TypeScript
• Infra & Deployment: Docker, NGINX, AWS (EC2/S3), CI/CD, Nosana (GPU Compute)

• AI Engineer Intern — Hooman Digital:

  • Built the backend infrastructure for decentralized GPU inference.
  • Managed container lifecycles and job scheduling systems.

• Generative AI Intern — TITAN Company:

  • Refactored legacy Flask services to FastAPI, boosting throughput by 2.5x.
  • Built workflows with LangGraph to slash manual cataloging time by 60%.

• AI R&D Intern — eBramha Techworks:

  • Research-focused backend work for NLP summarization and Speech-to-Text systems.

• Email: harishkb20205@gmail.com
• Portfolio: harish-kb.web.app
• LinkedIn: harish-kb-9417ba252
• X / Twitter: @harish20205
• Resume: View PDF

Open to collaborations on efficient LLM inference, model optimization, and distributed systems.