Skip to main content
Jayakrishna Konda - Data Scientist

Jayakrishna Konda — ML/AI Engineer

About Me

I'm Jayakrishna Konda — an ML/AI Engineer and Data Scientist with 5 years building production ML systems, RAG pipelines, and self-hosted AI infrastructure. I design end-to-end solutions: from model training and LLM fine-tuning to cloud deployment and real-time monitoring. When I'm not shipping ML features at work, I'm operating Batcave — my 56-container private AI server running LLMs, RAG, and ML inference 24/7.

Neural profileML/AI EngineeringGenAI & RAGMLOps + DevOpsSelf-Hosted AI

Production AI

5+ yrs

MLOps + DevOps

End-to-End

Batcave Uptime

99.9%

LLMs Self-Hosted

5+ Models

Experience Highlights

  • ▶5 years delivering production ML/AI systems at Cognizant, UMBC R/SEEK, and Enigma Technologies
  • ▶Expert in Python, deep learning, NLP, LangChain, RAG pipelines, and GenAI systems
  • ▶Built and operated Batcave — a solo-managed, 56-container private AI server at 99.9% uptime
  • ▶End-to-end cloud-native MLOps on AWS (SageMaker, Lambda, S3) and Oracle Cloud (OCI)
  • ▶Edge AI engineer: TensorFlow Lite on ESP32; custom Android ROM and kernel developer

Data Science Highlights

  • ●Production NLP pipelines across Legal, Healthcare, and Banking domains (Cognizant, 10M+ records)
  • ●LLM fine-tuning with LoRA/QLoRA; semantic caching for LLM cost reduction at Enigma Technologies
  • ●Wildfire detection: AllCNN on Sentinel-2 satellite imagery, 91% accuracy and 88% F1-score
  • ●Financial risk suite: FinBERT classifier (87% accuracy) + 1,000+ Monte Carlo simulations

Infrastructure & Edge Systems

  • ◆Repurposed Android device as low-power edge node with kernel-level CPU tuning (custom ROM)
  • ◆Built a fallback control plane: DDNS updater, Wake-on-LAN trigger, and Telegram webhook notifier
  • ◆Immich ML photo platform (401 GB, face recognition + object detection) and Paperless-NGX OCR
  • ◆177 GB multi-database data layer: PostgreSQL, MariaDB, Redis, SQLite, and Meilisearch

Featured Projects

  • ★Batcave: solo-built 56-container AI platform — LLMs, RAG, ML inference, 99.9% uptime (5-part blog series)
  • ★RAG Podcast Generator: PDF → LLM → TTS pipeline, 98% OCR accuracy, 4× throughput via LoRA fine-tuning
  • ★Autonomous RC car: YOLOv8 object detection (95% accuracy) + TFLite on ESP32 — no cloud dependency
  • ★Infrastructure Security Audit: automated scanner across 56 containers, identified 27 credential exposures
Home Server Live

Private infrastructure, public-safe telemetry

A sanitized view of my homelab operations: architecture, uptime signals, service categories, and telemetry patterns without exposing container names, ports, or internal endpoints.

Open telemetry
Live

Core Node

Private compute and storage

Live

Backup VPS

Edge and backup operations

Live

Telemetry

Live health snapshots

Live

Access Layer

Private mesh and identity

Smart Hub

Start where your last visit left off

Continue exploring

Once you explore a few pages, this space becomes your quick return rail.

Suggested collection

Full-Stack Work Path

A compact route through interface craft, end-to-end application design, and user-facing tooling.

Featured build

AI Platform Infrastructure — "Batcave" Personal ML Cloud

Solo-built private AI cloud on a Beelink mini PC — 56 containers, 36+ self-hosted services, 99.9% uptime. Runs local LLMs via Ollama, RAG pipelines, Authentik SSO, Tailscale VPN, and full Netdata monitoring.

Latest note

🧠 I Made My AI Ambush Me With Pop Quizzes

When an AI agent does the work, you risk learning nothing. So I wired Claude Code to randomly stop mid-session and quiz me on what we just did — turning passive assistance into active recall. Here's the build, and why Claude Code's hook system made it almost trivial.

9 min read

New since last visit

Global Light Mode Overrides Layer

Added a comprehensive theme-aware styling layer that automatically adapts cards, text, inputs, code snippets, hover states, selections, and interactive graphs to light mode, ensuring perfect contrast and readability across all pages.

AI Showcase now includes Podcasts

The /gallery page picked up a second section for AI-generated podcasts. Saved episodes show host tags, the TTS engine and Ollama model used, an inline audio player, and an expandable transcript. Save-to-showcase is gated by the same admin password as the image gallery.