ML & AI Enthusiast
Tejas Garg
Building ML Systems That Work Beyond Demos
Third-year undergrad • Seeking Summer 2026 ML/SWE internships
01. About Me
Building AI Systems
Third-year CS undergrad focused on building ML systems that work beyond demos. I care about the gap between research and production — making models interpretable, reliable, and deployable.
Recent work: reproduced a diffusion classifier from scratch and built custom XAI for it, created a real-time PPE monitoring pipeline with temporal filtering, and experimented with GRPO to teach LLMs explicit reasoning. I like projects where the engineering is as hard as the ML.
Education
B.Tech in Computer Science & Engineering
Specialization in AI & Machine Learning
Indian Institute of Information Technology, Nagpur
Expected 2027
Currently
- ▹Looking for Summer 2026 ML/SWE internships
- ▹Building interpretable LLM pipelines
- ▹Exploring RL for reasoning alignment
02. Skills & Technologies
Tech Stack
ML & Deep Learning
PyTorch & Lightning
Hugging Face Transformers
YOLO / Ultralytics
Diffusion Models
RL (GRPO, PPO)
Languages & Tools
Python
TypeScript/JavaScript
C++ / C
SQL
Git & GitHub
Web & Infrastructure
Next.js & React
FastAPI & Flask
Supabase & PostgreSQL
Docker
Vercel & AWS
Focus Areas
Explainable AI • LLM Pipelines & Evaluation • Reinforcement Learning for Reasoning • Diffusion Models • Real-time Computer Vision • Production ML Systems
Certifications
- ▹Deep Learning Specialization (DeepLearning.AI)
- ▹Machine Learning Specialization (DeepLearning.AI)
- ▹AWS Future AI Engineer (Udacity & AWS)
03. Featured Work
Projects
DiffMIC: Diffusion-Based Medical Image Classification
Reproduced DiffMIC-v2 from scratch — a dual-granularity conditional diffusion model for diabetic retinopathy grading. Matched the paper's 84.1% accuracy on APTOS 2019. Then built a custom XAI framework with 6 explainability techniques (temporal trajectories, attention maps, faithfulness validation) because traditional XAI doesn't work for 1000-step iterative inference.
PyTorchDiffusion ModelsXAIMedical Imaging
SentinelVision: Real-time PPE Compliance Monitor
Event-driven video processing system that turns noisy frame-level PPE detections into stable violation events. Built with YOLOv8/v11, SAM3, FastAPI, and Next.js. Handles real-world constraints: occlusion, limited GPU, long-running streams. Uses temporal filtering with EMA fusion and hysteresis thresholds to reduce alert spam.
YOLOv8FastAPINext.jsComputer Vision
NL-to-SQL: Interpretable Query Generation
Multi-stage LLM pipeline for reviewable SQL generation. Breaks down intent into reasoning steps, generates SQL, then validates and auto-corrects. Built-in security layer for prompt injection. Benchmarked on Spider dataset: 78% execution accuracy. Shows each decision so humans can audit before execution.
LLMFlaskQwenHuggingFace
GRPO: Teaching Mistral-7B to Reason
Experimented with Group Relative Policy Optimization to induce explicit reasoning in Mistral-7B via XML-structured traces. SFT warmup on GSM8K, then GRPO to encourage step-by-step thinking. Improved test accuracy from 41.2% to 52.5% (+11.3%). Learned critical lessons about evaluation consistency in RL for LLMs.
GRPOMistral-7BReasoningRL
BloodParser: AI Blood Test Dashboard
Full-stack medical analytics dashboard that extracts biomarkers from lab reports (PDF/images) using Gemini vision, visualizes them with interactive gauges, and provides an AI chatbot for questions about your results. Glassmorphic UI with health score aggregation.
Next.jsGeminiHealthcareTailwind