Satyabhan Umed Bhadoriya — Senior AI Engineer | Production GenAI, LLMs & Agents

01 Experience / 2018 — present

Independent AI Engineer (Self-Directed)

Pune, India

MAY 2024 — PRESENT

Lead active development and pilot management of CLINIRISE for 2 live clinic clients, continuously iterating on core RAG quality and prompt reliability.
Architected and deployed comprehensive open-source engineering curricula, building full-scale labs for eBPF observability, vLLM model serving, and AI threat defense.

Technical Support Engineer

Aman Security India Pvt. Ltd. — deployed on-site at CBSE Regional Office, Pune

MARCH 2024 — MAY 2026

AI Product Development: Architected and independently built CLINIRISE (outside of core support duties) applying GenAI, Python, RAG, and Vertex AI.
AI SDLC & Voice Prototyping: Utilized Claude Code and Antigravity IDE for an AI-assisted development workflow, setting up CI/CD pipelines via GitHub Actions. Prototyped STT/TTS pipelines for future voice-enabled patient interactions.
Infrastructure & Monitoring: Managed virtualized VMware environments, monitoring infrastructure health and service uptime using Prometheus and Grafana dashboards.
Technical Operations: Provided Tier-level technical support and issue resolution using ServiceNow, Jira, and remote access diagnostic tools.

Linux Administrator

Oceanic EOPS LLP — remote support for client sites, Pune

2023 — 2024

Managed RHEL 7 production servers, ensuring high availability and secure infrastructure.
Applied DISA STIG hardening, executed OpenSCAP compliance auditing, and configured strict SELinux policies.
Maintained continuous system visibility and monitored performance metrics via Grafana.

Application Engineer

Precision Automation, Pune

2018 — 2023

Delivered end-to-end automation engineering across ~15 major industrial projects, including assembly lines, packaging, and material handling systems.
Programmed advanced HMI/PLC logic (GX Works / Mitsubishi), consistently maintaining 99.8% operational uptime across deployed systems.

02 Featured project / shipped solo, end to end

CLINIRISE — Autonomous Healthcare CRM & RAG System

Multi-clinic WhatsApp CRM, solo-built end to end — live app ↗

● Pilot testing — 2 clinic clients

WhatsApp
Business API

→

Gemini LLM
+ Qdrant RAG

→

Express / Node
on Cloud Run

→

Firestore
+ Redis cache

→

Firebase
Hosting

Solo-built, end-to-end multi-clinic CRM currently in pilot testing with 2 live clients — view live app ↗

Multi-Agent System Design: Architected a supervisor-worker multi-agent topology using LangGraph-style orchestration, including intent-router, scheduling, FAQ RAG, and escalation agents. Implemented typed tool calls, shared agent state, and confidence-gated handoffs for clean agent-to-agent delegation.
RAG Architecture & Implementation: Engineered hybrid retrieval via Qdrant vector search and two-tier Redis caching to achieve sub-2 second response latency. Built semantic intent routing to separate deterministic transactional actions from generative queries, reducing LLM token consumption by ~40%.
AI Model Optimization: Developed multi-tier model resiliency (Gemini Flash → Flash-Lite → template fallback) integrating INT8 quantization, KV-cache optimization, and continuous batching. Benchmarked LLM/STT/TTS stacks against TTFT, error drift, and token cost using structured LLM evaluations.
AgentOps & Responsible AI: Implemented clinical guardrails including PII redaction (DPDP Act compliance), consent management, and human handoff escalation. Applied TypeSafe Jev (Choice/Score/Boolean) for fast, low-cost verification in place of full LLM calls.
Cloud & DevOps: Built with React 18, TypeScript, and Vite frontend alongside a Node.js/Express backend on GCP Cloud Run. Managed infrastructure via Docker, GitHub Actions CI/CD, and Workload Identity Federation. Validated with 860+ automated tests (Vitest, Cypress E2E, k6) targeting a scale of 1,000 clinics processing 100 patients/day.

WhatsApp Business API Google Gemini Qdrant RAG GCP Cloud Run Express / Node.js React 18 + TypeScript Firestore Redis / Upstash Firebase Hosting BigQuery + Looker Vitest · Cypress · k6 GitHub Actions

03 Skills / what I actually use, daily

🧠 AI & Systems Engineering

LLM Deployment, GenAI, RAG Architecture, Tool/Function Calling, Prompt Engineering, and Multi-Agent Systems

🤖 Frameworks & Libraries

PyTorch, Hugging Face Transformers, FastAPI, LangChain, LangGraph, and TypeSafe Jev

🚀 AI Model Optimization

LoRA/DPO Fine-Tuning, INT8 Quantization, vLLM Serving, and KV-Cache Optimization

🎙️ AI Platforms & Inference

GCP Vertex AI (Gemini), Groq (Fast Inference), Ollama, Azure STT, and ElevenLabs

🛠️ AI SDLC Tools

Claude Code, GitHub Copilot, and Antigravity IDE

🗃️ Vector & Data

Qdrant (Vector DB), Redis, PostgreSQL, and Firebase/Firestore

☁️ Cloud, DevOps & Infra

Linux/RHEL 7, Docker, Terraform, Ansible, Prometheus, Grafana, GitHub Actions, VMware, OpenSCAP, and STIG Hardening

💻 Languages

Python (Primary), TypeScript, and Node.js

04 Open-source engineering / 400+ chapters, CI-validated

Author & maintainer of production-depth engineering curricula and interview playbooks — each its own live site with runnable labs and automated CI. Built outside client work, in the open, continuously growing.

🚀 INTERVIEW DEBRIEFS · COURSE ● 11 Modules · 50 Scenarios

AI & DevOps Mastery

50 senior interview scenarios deconstructed: Reasoning Models (o1/R1), Triton kernels, vLLM serving, LangGraph bugs, 50M vector memory math, and 45-minute timed interview drills.

Explore the masterclass →

Timed Whiteboard Rehearsal · Interactive Calculators · Signal Rubrics

🧠 AI / ML ENGINEERING ● 9 parts · 138 ch

AI-ML

From-scratch autograd, BPE tokenizers, scaled dot-product attention and embeddings — then production system design, deployment, observability and enterprise-scale AI. Parts 1–5 build the primitives in pure Python; Parts 6–9 operate the standard production stack.

Explore the curriculum →

69/69 automated CI checks · zero dependencies · 6 live browser labs

☁️ CLOUD & DEVOPS ● 5 parts · 88 ch

DevOps-Infra

Infrastructure as Code, Terraform, Docker and Kubernetes from first principles — then IaC at scale, platform engineering, cloud networking & identity, and observability & incident engineering. Runnable labs, 3 a.m. production scenarios and SRE playbooks.

Explore the curriculum →

4-gate CI (validate + terraform + docker + kubeconform) · 14 deep-dive guides · 20+ lab files

🛡️ LINUX, HPC & SECURITY ● 21 modules · 126 lessons

linux-hpc-security

Linux internals and DISA STIG hardening, HPC benchmarking, eBPF, Ansible, Warewulf, Slurm, RDMA/InfiniBand fabric, parallel storage, kernel tuning, SOC/NOC operations, Splunk SIEM and red/blue/purple teaming — modelled in runnable pure-Python labs.

Explore the curriculum →

63 runnable labs · 800+ interview Q&A · 5 course parts

🔒 AI THREAT DEFENSE ● Production Grade

AI Threat Defense

Production-grade AI Threat Detection, Adversarial Robustness & Go Gateway Engineering Curriculum.

Explore the curriculum →

Live Curriculum

★ REPOSITORIES & VERIFIED CI

The FDE Playbook: Learning AI vs Engineering AI — interactive state diagrams, ELI5 analogies, and click-to-run simulators across 10 production interview topics.
AI & DevOps Mastery: 11-module interactive course deconstructing 50 senior AI & DevOps interview scenarios with live calculators and signal rubrics.
AI-ML & AI-Lab: Authored a 9-course curriculum covering scalar autograd, BPE tokenizers, attention mechanisms, RAG + LangGraph/ReAct agent loops, LoRA/DPO fine-tuning, INT8 quantization, KV-cache, and vLLM serving. Maintains 69/69 passing CI checks.
AI Threat Defense: Developed production-grade training materials for AI Threat Detection, Adversarial Robustness, and Go Gateway Engineering.
DevOps-Infra: Built extensive SRE playbooks and infrastructure code utilizing Terraform IaC and Docker multi-stage hardened builds, covering SLI/SLOs, golden signals, and incident runbooks secured by a 4-gate CI pipeline.
Linux-HPC-Security: Created comprehensive systems engineering modules from the metal up, detailing Linux internals, hardening, DISA STIGs, Ansible, Slurm HPC, and featuring an advanced eBPF Observability Capstone.