✦ Based in Jaipur, Rajasthan

JAI DHINGRA

Machine Learning Engineer — GenAI Systems

Designing systems where data becomes intelligence — building GenAI architectures that retrieve, reason, and respond in real time. |

View Architecture & Projects Download Resume

Production Focus

Leading GenAI Platform (DOD)

Humanli.AI

Performance Win

41% Latency Reduction

Inference Optimisation

Architecture

Multi-Agent RAG

Deployed on Azure

scroll

✦ Scene II

AI Memory
Architecture

A system where active intelligence is cached and experience is permanently indexed.

Short-Term Memory

Redis Cache :: Fast Retrieval

Active Skills

🧠 Core Intelligence

›Designing RAG pipelines with contextual retrieval and citation grounding

›Building multi-agent systems with tool-based reasoning and routing

›Structuring LLM workflows for real-world decision systems

⚡ Optimization & Modeling

›Fine-tuning Gemma (270M / 1B) using LoRA for domain-specific QA

›Reducing inference latency (41% improvement) via load balancing & batching

›Embedding strategies for high-quality semantic retrieval

🗄️ Memory & Retrieval Systems

›Vector search using FAISS / ChromaDB

›Redis → short-term contextual memory

›PostgreSQL → persistent structured memory

›Chunking, indexing, and retrieval optimization for documents

☁️ Systems & Deployment

›Deploying LLM systems on Azure VM with Docker

›CI/CD pipelines using Azure DevOps

›Building scalable APIs with FastAPI

🔬 ML Foundations

›PyTorch (model building & inference)

›TensorFlow & Scikit-Learn (classical ML pipelines)

$ redis-cli GET active_skills

{

domain: "Applied GenAI Systems",

focus: ["RAG", "Agents", "LLM Optimization"],

infra: ["Azure", "Docker", "CI/CD"]

state: "Continuously evolving",

}

Long-Term Memory

PostgreSQL :: Persistent Store

Career Timeline

May 2025 – Present

Humanli.AI

Machine Learning Engineer

★ Architecting a Data-on-Demand GenAI platform with multi-agent RAG systems and optimized inference pipelines

›Leading development of enterprise-grade GenAI platform for intelligent data retrieval

›Built multi-agent workflows for query routing and document reasoning

›Designed full RAG pipeline with LangChain, FAISS, and Mistral/Gemma

›Implemented OCR ingestion pipeline for PDFs/images with chunking + embeddings

›Fine-tuned Gemma models using LoRA for domain-specific QA

›Reduced inference latency by 41% via system optimization

›Designed hybrid memory architecture (Redis + PostgreSQL)

›Deployed scalable systems using Azure VM, Docker, and CI/CD pipelines

June 2024– August 2024

Semi-Hyphen

ML Engineer Intern

○ Built real-time Yoga Pose Detection system for mobile applications

›Developed pose detection pipeline using MediaPipe + MobileNetV2

›Achieved 88% accuracy across 8 yoga poses

›Focused on real-time inference and mobile-friendly performance

›Strengthened understanding of computer vision pipelines

⚡Deployed on Azure VM via Docker containers

✦ Scene III

Project
Case Studies

User QueryInputRouter NodeLangGraphClinical AgentFAISS · PDFsReceptionist AgentGemini APIOutconditional routing & escalation

Scene III · 01

Post-Discharge Medical AI Assistant

Designed a multi-agent GenAI system for post-discharge patient support, routing queries through a clinical knowledge base with intelligent fallback. The system combines structured medical retrieval with real-world adaptability to ensure reliable responses.

  • •Built LangGraph router for query classification and agent orchestration
  • •Implemented RAG over nephrology PDFs using FAISS with citation grounding
  • •Integrated web fallback for out-of-distribution queries
PythonLangGraphFastAPIFAISSAzure
30 FPS ✓Flask APIRESTDockerContainerEdge ⚡

Scene III · 02

FastSAM Real-Time Inference

Engineered a real-time instance segmentation system optimized for edge deployment, enabling live video understanding with minimal latency.

  • •Achieved ~30 FPS through optimized inference pipeline
  • •Packaged as Flask microservice inside Docker container
  • •Designed for plug-and-play integration into downstream systems
PyTorchFastSAMFlaskDockerOpenCV
LIVEPhi-3 Mini

Scene III · 03

AI Call Assistant

Built a low-latency voice-to-LLM pipeline enabling real-time conversational AI without external API dependency.

  • •Streamed audio via Web Speech API with real-time transcription
  • •Integrated Phi-3 Mini via Ollama for local inference
  • •Achieved sub-500 ms round-trip latency
Web Speech APIFastAPIPhi-3OllamaWebSockets

✦ Scene IV

Deployment
Pipeline & Stack

Production-grade CI/CD — code flows from commit to containerised microservice on Azure in minutes.

⊕

Git Commit

Source Control

⟳

Azure DevOps

CI/CD Pipeline

▣

Docker Build

Containerize

◈

Azure VM

Virtual Machine

✓

Production API

Live Endpoint

Cloud

Azure VMAzure DevOpsAzure Blob

Containers

DockerDocker Compose

ML Frameworks

PyTorchLangChainFAISSTransformers

Databases

PostgreSQLMongoDBRedis

APIs & Infra

FastAPIFlaskGitCI/CD

✦ Let's Build

Let's Architect
Something.

Open to GenAI / ML engineering roles, GenAI consulting, and interesting hard problems. Based in Jaipur — available globally.

github.com/jaidh01linkedin.com/in/jaidhingrajaidhingra402@gmail.com
Download Resume

Jai Dhingra · Machine Learning Engineer — GenAI Systems · Jaipur, Rajasthan