Parth Shastri
CV
Illustrated portrait of Parth

Hi, I'm Parth.

I'm an engineer who got hooked on how machines learn, and never really got unhooked.

These days I build retrieval, agent & voice systems. On the side I'm building Theos (my own assistant) & co-founding bev.

I like taking a messy idea, working out the math, then shipping something clean.

Let's work together Download CV
App. B

Work history

  1. Jan 2026 - NowPune

    Technical Lead · Fulcrum Digital

    • Leading AI research & engineering for AI Studio, a RAG-based knowledge platform for insurance, e-commerce & enterprise.
    • Teams ingest policies, contracts, SOPs & CRM data, then ask questions over chat, voice or API.
  2. Aug 2025 - Dec 2025Pune

    Senior AI Scientist · Fulcrum Digital

    • Built an invoice platform for 10,000+ documents with 95%+ structured extraction accuracy.
    • Built a ReAct agent with RAG over PostgreSQL, streaming its first token in under 100ms.
    • Built a SIP voice agent with under 500ms voice-to-voice latency & 99%+ uptime.
  3. Sep 2023 - 2025Hyderabad

    Data Scientist · Bridgeweave

    • Built deep learning models with evolutionary search & computer vision to pick stock baskets (50%+ annual returns, 100 Cr+ AUM).
    • Fine-tuned an LLM agent for stock analysis & user education, with guardrails that cut hallucinations by 60%.
  4. Jun 2022 - Aug 2023Pune

    Machine Learning Engineer · Searce

    • Shipped power production forecasting for Tata Power (95% accuracy, XGBoost on Vertex AI).
    • Built document intelligence for Apollo 24/7 & Rockwell Automation, including a fine-tuned CascadeTabNet with 98% table accuracy.
App. C

Papers

Google Scholar →
  1. [1]
    D²Styler: Advancing arbitrary style transfer with discrete diffusion methods

    O. Susladkar, G. Deshmukh, S. Mittal, P. Shastri. International Conference on Pattern Recognition (ICPR 2024), Springer, 2025. arXiv

  2. [2]
    Adversarial synthesis based data augmentation for speech classification

    P. Shastri, C. Patil, P. Wanere, S. Mahajan, A. Bhatt. International Conference on Signal and Information Processing (IConSIP), IEEE, 2022.

  3. [3]
    Adversarial synthesis based data-augmentation for code-switched spoken language identification

    P. Shastri, C. Patil, P. Wanere, S. Mahajan, A. Bhatt, H. Sailor. arXiv:2205.15747, 2022. Code

Want to build something together?

Collaborations, research chats, podcast guests & interesting side projects. I'd love to hear from you.

Get in touch