Hi, I'm

Parth Srivastava

AI/ML Engineer

I build intelligent systems that learn, reason, and solve real-world problems.

Explore My Work Download Resume
Python FastAPI AI Agents Recursive Planning Server-Sent Events (SSE) PostgreSQL MongoDB Atlas AES-256-GCM Vault RunPod Novita AI Vast.ai LangChain LangGraph PyTorch Hugging Face vLLM Docker SQL KQL

Professional Experience

AI/ML Engineer

4Way Technologies
Nov 2025 – Present
  • High-Throughput Model Serving: Deployed 20+ AI models on cloud GPU platforms (RunPod, Novita AI) using Docker, vLLM, and FastAPI.
  • Medallion Data Lake ETL: Built Apache Airflow ETL pipelines following Bronze, Silver, and Gold data lake architecture, processing 10K+ records daily.
  • Real-Time Dynamic RAG: Built a RAG-based chatbot with real-time knowledge synchronization, supporting dynamic document ingestion, live updates, and deletions.
  • Multilingual AI Assistant: Developed a multilingual AI assistant integrated with RAG pipelines for context-aware conversations across multiple languages.
  • Observability & Load Testing: Performed API load testing using Locust and monitored latency, throughput (RPS), error rates, and concurrent user performance with Prometheus.
  • Generative Media Workflows: Developed automated image generation workflows using Stable Diffusion, ComfyUI, and Automatic1111, serving 200+ active users.

Featured Projects

View all repositories
RunPod Novita AI Vast.ai MODEVA ENGINE LIFO Stack Planner SSE Stream + Vault 1-Click Safety Cards Spot Pricing $0.19/hr (4090) Persistent Vol Datacenter Sync Live Orchestrator: RunPod · Novita · Vast.ai Live Online ⚡
Multi-Cloud Infrastructure & Autonomous AI Orchestration

Modeva — Autonomous Multi-Cloud GPU Orchestrator

Unified multi-cloud control plane that aggregates real-time GPU spot pricing, datacenter inventory, and persistent storage across RunPod, Novita AI, and Vast.ai. Features an autonomous conversational assistant with recursive backward-chaining stack planning and real-time Server-Sent Events (SSE) telemetry.

Python FastAPI PostgreSQL MongoDB Atlas LangChain Groq SSE AES-256-GCM Docker SQLAlchemy
Qwen2.5 QLoRA Fine-Tuning Loss Label Mask (-100)
LLM Fine-Tuning & SFT

Qwen2.5 LLM Fine-Tuning Pipeline

Fine-tuned Qwen2.5-0.5B using Full Fine-Tuning and QLoRA on RunPod GPUs for structured JSON extraction. Includes 10K+ synthetic instruction dataset and custom label masking (-100).

PyTorch HuggingFace PEFT QLoRA vLLM RunPod
View HuggingFace Model
LangGraph SDV TVAE Dataset Outlier Preservation • Pydantic Schema Validated
Autonomous Data Workflows

GenData - Synthetic Data Generator

Multi-agent synthetic dataset creation tool driven by LangGraph workflows, Groq LLaMA 70B, and SDV TVAE tabular synthesis with automated EDA and Pydantic validation.

Python LangGraph Groq LLaMA 70B SDV TVAE Pydantic
View GitHub Repository
STT (Sarvam) LLaMA 3.3 70B Agent TTS Synthesis Deployed Railway
Real-Time Voice AI & Deployment

Say - Voice-to-Voice AI Agent

End-to-end real-time voice-to-voice conversational agent with STT, Groq LLaMA 3.3 70B, and TTS synthesis. Fully containerized with Docker and deployed on Railway.

FastAPI Docker Groq LLaMA 3.3 SarvamAI Railway
View GitHub Repository
cmd - bitfy cli C:\> bitfy --ask "get system memory" > FAISS Vector Search: matched query Proposed: wmic os get freephysicalmemory Execute command? [y/N]: y
Developer Tooling & CLI

Butterfly (Bitfy CLI)

AI-powered command-line assistant converting natural language into safe system commands. Features persistent local FAISS vector memory store and async threading.

Python Click LangChain FAISS Sentence-Transformers
View GitHub Repository

Core Skills & Stack

AI & LLM Systems

AI Agents Recursive Stack Planning Multi-Turn RAG LLM Fine-Tuning PEFT & QLoRA Synthetic Data (SDV) NLP Deep Learning

Frameworks & Backend

FastAPI LangChain LangGraph Server-Sent Events (SSE) SQLAlchemy Pydantic PyTorch Hugging Face Transformers FAISS Scikit-learn XGBoost LightGBM

Databases, Security & Cloud

PostgreSQL MongoDB Atlas AES-256-GCM Cryptography Clerk / JWT Auth Docker RunPod Novita AI Vast.ai vLLM Linux Apache Airflow Prometheus Locust Git & GitHub

Programming

Python SQL KQL (Kusto Query Language)
Currently
Exploring
LLMs & Agents
Building agentic systems that can reason and act.
RAG Systems
Improving retrieval, re-ranking and context utilization.
NLP Research
Exploring long-context understanding and generation.
Open Source
Contributing to tools and libraries for the community.

About Me

Parth Srivastava

I'm an AI/ML engineer with a strong interest in Natural Language Processing, Information Retrieval, and building practical AI systems. I love turning research ideas into real-world applications.

Currently working at 4Way Technologies, where I deploy production LLM services using Docker & vLLM, design Airflow Medallion ETL pipelines processing 10K+ daily records, and build real-time vector synchronized RAG applications.

When I'm not coding or training models, I'm either reading research papers, lifting, or listening to music.

Know more about me
1+
Years of Experience
20+
AI Models Deployed
10+
Projects Built

Interactive AI Playground

Live Inference Simulation
Input Unstructured Prompt
Model Output (Token Masked Response)
// Click "Extract Structured JSON" to simulate model inference...