█████╗ ██╗ ██╗██╗██╗ ██╗██╗ ██╗███╗ ███╗ █████╗ ██████╗
██╔══██╗██║ ██║██║██║ ██╔╝██║ ██║████╗ ████║██╔══██╗██╔══██╗
███████║██║ ██║██║█████╔╝ ██║ ██║██╔████╔██║███████║██████╔╝
██╔══██║╚██╗ ██╔╝██║██╔═██╗ ██║ ██║██║╚██╔╝██║██╔══██║██╔══██╗
██║ ██║ ╚████╔╝ ██║██║ ██╗╚██████╔╝██║ ╚═╝ ██║██║ ██║██║ ██║
╚═╝ ╚═╝ ╚═══╝ ╚═╝╚═╝ ╚═╝ ╚═════╝ ╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝
Building production-grade AI systems — from RAG pipelines to agentic workflows
class Avikumar:
name = "Avikumar Talaviya"
role = "AI Engineer & Data Scientist"
focus = ["Agentic AI Systems", "RAG Pipelines", "Medical AI", "LLM Apps","Data pipelines","Data Analytics"]
building = "Production-grade AI backends with FastAPI + LangGraph + Cerebras"
open_for = ["Collaborations", "Open Source", "Freelance AI Projects","Full time roles"]I'm a hands-on AI engineer who loves turning cutting-edge research into working systems. From multi-agent RAG architectures to medical image-to-text models, I build end-to-end — backend, inference, and deployment.
AI / ML
Backend & APIs
LLMs & Inference
Agentic RAG system for veterinary knowledge
FastAPI · LangGraph · Cerebras · Redis (two-tier) · Sentence-Transformers
A production-grade multi-agent RAG backend with ephemeral + persistent memory tiers, local embeddings, and blazing-fast Cerebras LLM inference. Built for real-world veterinary protocol retrieval.
AI-powered career intelligence platform
Next.js · FastAPI · Cerebras SDK · Supabase
Generates match scores, skill gap analysis, and AI-rewritten resumes tailored to job descriptions — full-stack, cloud-ready, and production deployed.
BiomedCLIP + Clinical-T5 with Visual Abstractor
PyTorch · LoRA · Gated Cross-Attention · MIMIC-CXR · AMP · EMA
A hybrid vision-language model architecture for automated radiology report generation. Trained on MIMIC-CXR with sinusoidal 2D positional encodings, LoRA fine-tuning, and ROUGE-L validation monitoring.
End-to-end LLM learning hub
Covers fine-tuning (LoRA, PEFT, PPO, DPO), semantic search, RAG with ChromaDB, Transformer internals, and deployment. A living resource for engineers mastering the full LLM lifecycle.
Statistical forecasting on real-world datasets
Statsmodels · S&P 500 Analysis · ARIMA · Prophet
Top-10 S&P 500 stocks analysed and forecasted using classical + modern time-series methods.
Multiclass classification on highly imbalanced real-world data
Scikit-learn · SMOTE · EDA · Addis Ababa Police Dataset
Solved a real-world 3-class imbalance problem using resampling strategies and ensemble classifiers.
- 🤖 Multi-agent orchestration with LangGraph & custom tool loops
- ⚡ Cerebras ultra-fast inference for real-time AI applications
- 🏥 Medical AI — vision-language models for clinical use cases
- 🔍 Advanced RAG — hybrid retrieval, re-ranking, and query routing
"Build fast. Break things. Learn everything."