Shaktisinh Chavda
AI/ML Engineer • Full Stack AI Developer • LLM & RL Specialist
AI isn't the destination, Autonomy is!
Building Intelligent Systems with Purpose
AI/ML Engineer with hands-on experience in LLM fine-tuning (SFT, LoRA, QLoRA, GRPO), Reinforcement Learning, Retrieval-Augmented Generation (RAG), and multi-agent system design. Skilled in building end-to-end ML pipelines — from model training and optimization to production deployment using FastAPI, Docker, and REST APIs.
Shaktisinh Chavda
B.Tech, AI & Machine Learning
LD College of Engineering · Ahmedabad
Technical Expertise
Core technologies and frameworks I work with
Languages
ML / Deep Learning
AI / LLM / GenAI
Tools / Infrastructure
Featured Projects
A selection of AI & ML projects I've built
Fine-tuned Qwen2.5-0.5B for autonomous web navigation using a two-phase pipeline (SFT warm-up → GRPO), achieving 100% task accuracy on BrowserGym benchmarks — 4× improvement over zero-shot baseline. Dockerized end-to-end RL training pipeline with custom reward shaping, optimized for a single 4 GB VRAM GPU.
Production-ready AI web editor that translates natural language instructions into real-time UI modifications using Gemini Flash and local LLMs, processing 50+ concurrent edit sessions. Features a non-destructive JSON style-patching engine with intelligent element bubbling, preserving complex DOM structures.
AI-powered data analytics platform enabling natural language querying, manipulation, and visualization via a multi-agent LangChain architecture — 40% faster than single-agent baseline. Integrated local LLM inference (Ollama) for fully offline, privacy-preserving enterprise data analysis with automated chart generation.
Deep learning framework for manipulated media detection using domain-adversarial training, improving cross-domain accuracy by 15% on unseen data distributions. Curated 10,000+ sample dataset spanning FaceSwap, Face2Face, and NeuralTextures generation techniques for robust evaluation.
Experience
Professional journey in AI & Machine Learning
- Optimized ML inference pipeline using model quantization and batch processing, reducing API response latency by 20% and improving throughput for 10K+ daily prediction requests.
- Deployed real-time model monitoring dashboard (Python, FastAPI) with automated prediction drift detection, reducing production bug identification time by 35%.
- Shipped 3 ML-driven features to production within a 4-week sprint cycle, collaborating with cross-functional engineering and product teams using Agile methodology.
Education
Academic foundation in AI & Machine Learning
B.Tech, Artificial Intelligence & Machine Learning
L.D. College of Engineering, Ahmedabad
Relevant Coursework
Achievements
Key milestones and accomplishments
100% BrowserGym Benchmark
Achieved state-of-the-art accuracy with a 0.5B-parameter model on autonomous web navigation tasks.
Deepfake Research Breakthrough
Built domain-adversarial detection framework with 15% cross-domain accuracy gain; 10K+ curated evaluation samples.
Open Source Contributor
Maintained 6+ public AI/ML repositories on GitHub with production-quality documentation.