DIVAX SHAH
AI & ML ENGINEER.
Building AI applications powered by fine-tuned LLMs, generative models, and NLP. Focused on making AI useful, reliable, and deployable.
EXPERIENCE.
01 - TIMELINEAVINYAA EDTECH
JR. AI/ML DEVELOPERArchitecting advanced grammar checkers and fine-tuning LLMs for pristine natural language output. Designing multilingual AI classification systems with 95%+ accuracy.
THINKBIZ TECH
JR. PYTHON DEVELOPERConstructed scalable chatbot PoCs utilizing 17+ LangChain Agents with structured/unstructured ingestion. Engineered OCR-LLM pipelines improving invoice parsing from 15% to 85% accuracy.
DMI FINANCE
AI & SYNTH DATA INTERNEngineered generative pipelines for synthetic structured data via Gradio & Python. Leveraged PyTorch to dramatically increase dataset throughput and deduplication algorithms.
PROJECTS.
02 - ARCHITECTURAL SYSTEMSLLMS & AI ARCHITECTURES
SANSKRIT QWEN2.5-7B CHAT
Specialized language model for Sanskrit translation and transliteration. 100% success rate on test sequences.
SANSKRIT QWEN2.5-VL OCR
Vision-Language model adapted for Sanskrit OCR (Optical Character Recognition) tasks.
GITAWHISPER (WHISPER TINY)
Fine-tuned Whisper-tiny for Sanskrit shloka transcription with IAST transliteration. 35% WER reduction.
CUSTOM GPT 100M MODEL
GPT-style Transformer built from scratch on Fineweb via DeepSpeed, ZeRO Stage-2, and FP16 precision.
SANSKRIT TOKENIZER
Native tokenization offering 4.5x better efficiency over byte-level tokens, 120K vocab size.
GENERATIVE & SIMULATIONS
FLUX LORAS
Specialized generative AI adapters fine-tuned on diverse image datasets for aesthetic scaling.
LORE KEEPER
Infinite structural narrative generator using custom LLM agents and creative generation constraints.
WORLD SIM
CLI-native environment simulator allowing users to craft dynamic sandbox simulations via LLMs.