AI/ML Engineer @ 4Way Technologies

Engineering Production LLMs, RAG & Autonomous Agents

Specialized in fine-tuning open-source LLMs (Qwen, LLaMA), building enterprise RAG architectures with dynamic vector sync, synthetic data generation pipelines, and deploying high-throughput AI models via Docker & vLLM.

Launch AI Simulator View Projects
20+ AI Models Deployed
10K+ Daily ETL Records
200+ Image Gen Users Served
bash - parth@ai-workstation: ~/portfolio
Interactive
parth@ai-workstation:~$ initializing AI profile environment...
[OK] PyTorch, vLLM, LangGraph & Docker ready.
parth@ai-workstation:~$ Click any quick command chip below to run real-time CLI actions:
Technical Competencies

Core Skills & Stack

A comprehensive breakdown of frameworks, models, data pipelines, and infrastructure tools.

LLM Fine-Tuning & SFT
QLoRA PEFT Token Masking (-100) Hugging Face Synthetic Instruction Datasets
RAG & AI Agents
LangChain LangGraph FAISS Real-Time Knowledge Sync Multilingual RAG
High-Throughput Serving
vLLM RunPod Novita AI Docker FastAPI Linux
Data Engineering & ETL
Apache Airflow Bronze/Silver/Gold Lakehouse MongoDB Axiom (AQL) SDV TVAE
Observability & Load Testing
Prometheus Locust (RPS / Latency) ComfyUI Stable Diffusion Automatic1111
Programming Languages
Python (PyTorch, Scikit-learn) SQL KQL (Kusto Query Language) Pydantic
Career History

Professional Experience

Proven track record in deploying production AI models, building scalable ETL pipelines, and optimizing LLM serving.

AI/ML Engineer

4Way Technologies
Nov 2025 – Present
  • Model Deployment & vLLM Serving: Containerized and deployed 20+ AI models on cloud GPU platforms (RunPod, Novita AI) utilizing Docker, vLLM, and high-performance FastAPI wrapper services.
  • Data Lake ETL Architecture: Built enterprise Apache Airflow ETL pipelines following Medallion (Bronze, Silver, Gold) Data Lake architecture, reliably processing over 10K+ records daily.
  • Real-Time Dynamic RAG Chatbot: Architected a RAG chatbot supporting real-time knowledge vector synchronization, permitting dynamic document ingestion, live updates, and deletions on the fly.
  • Multilingual AI Assistant: Developed a context-aware multilingual assistant integrated into RAG pipelines for seamless cross-language user support.
  • Load Testing & Telemetry: Executed API load testing via Locust, continuously monitoring RPS throughput, latency distribution, error rates, and concurrent user performance using Prometheus.
  • Generative Media Workflows: Engineered Stable Diffusion, ComfyUI, and Automatic1111 automated image generation pipelines, serving 200+ active users.
Featured Work

Open-Source & AI Projects

Deep dive into custom fine-tuned LLMs, synthetic data generators, and AI agent architectures.

HuggingFace Fine-Tune

LLM Fine-Tuning Pipeline (Qwen2.5)

End-to-end QLoRA and Full Fine-Tuning pipeline targeting Qwen2.5-0.5B on RunPod GPU instances for structured JSON extraction.

  • Created a synthetic instruction dataset of 10K+ training samples.
  • Custom tokenization & prompt label masking (-100) for strict response learning.
  • Iterative progressive dataset cleaning stages (P0, P1) to resolve conflicting samples.
PyTorch Transformers PEFT QLoRA RunPod vLLM
View on Hugging Face
GitHub Project

GenData - Synthetic Data Generator

Autonomous multi-agent synthetic dataset creation tool driven by LangGraph workflows, Groq LLaMA 70B, and SDV TVAE tabular synthesis.

  • Generates high-fidelity datasets from raw single/multi CSV files.
  • SDV TVAE with validation & automatic retry logic to preserve statistical outliers.
  • Automated EDA extracts statistical correlations to direct synthetic dataset creation.
  • Pydantic schema validation enforcing field constraints, types, and diversity.
LangGraph Groq LLaMA 70B SDV TVAE Pydantic Python
GitHub Source Code
Live Demonstration

Interactive AI Simulator

Test Parth's Fine-Tuned Qwen2.5 JSON Extractor & LangGraph Synthetic Data Generator in real-time.

Input Unstructured Text
Model Output (Token Masked Response)
// Click "Extract Structured JSON" to simulate model execution...
Background & Recognition

Education & Achievements

Student Leader - Smart India Hackathon (SIH 2025)

Led the team in Smart India Hackathon 2025, driving technical architecture, ML solution design, and core decision-making.

B.Tech in CS & Engineering (Data Science)

Raj Kumar Goel Institute of Technology

Aug 2022 – Sep 2026 | Ghaziabad, Uttar Pradesh
CGPA: 7.4