Mihir Kothari
AI engineer. I build agent systems, vision pipelines and the products around them.
- Now
- AI Builder, Pravi Research
- Studying
- M.Tech in Data Science, Nirma University
- Based in
- Ahmedabad, India
In short
- Ships production software for public-sector programmes: live video analytics, field-workflow platforms, large records portals. My work at Pravi Research
- Won the Infineon Technologies hackathon at Nirma University, Feb 2026. Defend the Kingdom
- Built Reform.ai: research PDF to journal-formatted LaTeX with a single model pass. Reform.ai
- Founded Intelli-Gen on a state startup grant; generated 2100+ synthetic defect images during a pharma-inspection internship. Intelli-Gen
- Research with numbers: 0.84% EER audio deepfake detection that explains its decisions; mixed-precision LLM quantization. Two-Box Audio Deepfake Detection
Experience
AI Builder, Pravi Research
Jun 2026 → now
Production systems for public-sector programmes: video analytics, field-workflow platforms and large records portals. Details are confidential.
Founder, Intelli-Gen
Oct 2024 → now
My first startup: synthetic image datasets from a handful of reference photos. Funded under Gujarat’s Student Startup and Innovation Policy, presented at the NAMO iHub inauguration, and written up as a paper.
Generative AI Intern, SPAN Inspection Systems
May → Aug 2024
Led prompt-driven generation of defective blister-pack images with fine-tuned Stable Diffusion, for machine-vision quality inspection. 2100+ images delivered.
Education
M.Tech, Computer Science & Engineering (Data Science)
Nirma University, Ahmedabad, 2025 → 2027
B.E., Information Technology
LDRP Institute of Technology & Research, Kadi Sarva Vishwavidyalaya, 2021 → 2025. 8.99 / 10, First Class with Distinction
What I do
Agent systems
LangGraph, CrewAI, MCP, message-bus orchestration, structured outputs
Vision and video
Detection, tracking, plate reading, multi-stream pipelines on GPU
Generative imaging
Stable Diffusion, DreamBooth, fine-tuning for synthetic data
Model efficiency and evaluation
Mixed-precision quantization, benchmark design, honest baselines
Audio and interpretability
Anti-spoofing, graph attention, explanations tested for faithfulness
Product engineering
React, React Native, Node, FastAPI, PostgreSQL, release pipelines
Algorithms and simulation
Graph search, event-driven simulation, swarm coordination