Surya Pradeep Kumar Varma

Novartis | Gen AI Solutions, Innovation & Scalable Delivery
* Orchestrated the development of production-grade Gen AI solutions, aligning technical teams and
business stakeholders to deliver high-impact initiatives, driving wide adoption.
* Custom Content Generation Framework: Engineered a RAG pipeline (Llama Index) that reduced retrieval latency by 35% through holistic context integration, earning stakeholder endorsement for a PoC now in active development.
* Accelerated innovation by building 10+ scalable prototypes (e.g., Virtual Presenter Bot, Deep Research Agent) with modular architectures, reducing validation timelines by 40% and securing executive buy-in for R&D expansion.

Client Pitches & Demos
* Led creation of tailored Gen AI use-case decks for executives at the GDS Healthcare Conference, collaborating with data science, marketing, and design teams.
* Delivered technical pitches to 15+ US healthcare leaders, sparking partnership dialogues with Hackensack Meridian, Cano Health, and others.

Interviews & Hiring Processes
* Revamped Data Science hiring process, reducing mis-hire risk and time-to-hire via comprehensive candidate scoring and cross-functional interview panels. Interviewed numerous candidates to onboard to the Data Science and Generative AI Team.

Awards & Recognition
* Spot Award (Space Inventive): Recognized for accelerating Gen AI PoC development and championing cross-functional collaboration on technical architecture, showcasing innovation and leadership.

• Company Overview: Bangalore (Website)
• Led key projects at Curl Hg, taking on multiple responsibilities, driving the development of, and later took full ownership of multiple critical ML components for Sara Document AI :(Document Processing / Information Extraction Engine): NER Models (SPV, ANA, Banking NER), Address Extraction Model, Topology (Document Layout Parser), and a custom Post-processing engine with 95% test coverage in all the modules.
• Fine tuned BLOOM (BigScience Large Open-Science Open-Access Multilingual language Model) LLM as a Question Answering Model.
• Built and scaled custom NER engine forming the NER backbone for the product dealing with 19 different entities for information extraction from unstructured document data (multiple formats).
• Model is an ensemble of custom-built transformer based NER model, with a custom built dynamic searching algorithm and pattern-based heuristics followed by NER Entity conflict resolution to predict the custom NER entities.
• Profiled and Optimized the end-to-end pipeline in Sara-Backend to attain a mean latency improvement of 35% per page processed.
• Bangalore (Website)

• Company Overview: Bangalore (Website)
• Researched multiple NER models including flair, bert and spaCy.
• Developed Bi-LSTM based ANA (Anchor / keyword recognition NER model) with improved recall of 93%.
• Topology Classification ran 3x faster using the integrated ANA NER model.
• Bangalore (Website)

• Company Overview: Remote (Website)
• Built a Sentiment Classifier Model to classify sentiments from IMDB's textual movie reviews using traditional and deep learning techniques using Tensor flow, Keras, PyTorch and Bert libraries.
• Remote (Website)

• Company Overview: ITS Engg. College, Noida
• Developed three projects on a CNN-based digit classifier model, home automation, simulation of an Arduino on Simulide, NI LabVIEW and ThingspeakIOT platforms.
• ITS Engg. College, Noida

• Company Overview: Govt. Of India
• Worked on MietY Startup Hub(MSH) Project in the role of Trainee, worked on exploratory data analysis of data pertaining to diverse startups.
• Govt. Of India