MY STORY

Who am I?

Hi! I’m Varshit.

I am an AI consultant helping early stage start-up get started with their machine learning  program.

Currently, I am working as an AI consultant to Pfizer helping their commercial analytics team to improve revenue outcomes by building a full stack machine learning pipeline.

Before that I was consulting Enlyft's data science to build a natural language based product to boost the success rate of sales teams of top technology companies in the world such as Microsoft, Dell, Salesforce etc.

Until December 2023, I was the Head of Technology at Blackboard Radio - an Edtech start-up with the mission to support working professionals in improving their communication skill sets.

Previous to that I was advising Insciter, an early stage healthcare start-up based out of London to build a network graph of world's leading scientists and doctors and apply machine learning algorithm to help commercial teams of european pharma companies improve their revenue outcomes.

I have a strong education background in coding, healthcare, communications and data science. I completed my bachelor’s degree from IIT Kharagpur and earned a master's from Imperial College London.

During my time at IIT, I specialised in experimental bioengineering. I worked on vaccine development against a virus infecting Indian silkworm crop.

But at Imperial, I switched to the Maths department and got trained in computational biology (CS+Maths+Biology).

For 10 months, I worked on developing a new understanding of cancer metabolism. The particular focus of the project was comparing how cancer cells have a different metabolic than healthy cells and how that can be used to cure them.

I am also co-founder of India Biotech Leaders of Tomorrow (IBLoT) - a non-profit focussing on promoting Bio-entrepreneurship in India. I act as the communications for IBLoT and lead all strategic communication, Public Relations, and content marketing work.

Outside of work, I have a strong interest in performing arts.

I am an amateur theatre artist and performed in 12+ stage and street plays across India and London.

I am also passionate about public speaking. I am a communication consultant at Toastmasters International (a Global Non-Profit leading to improve Public Speaking) and have won several state-level Public Speaking contests.

I also have a strong interest in writing and have written for Felix (Imperial College Student Newspaper), as well as Oncobites (A cancer biology blog run by PhD students at UNC Chapel Hill), as well as manage the IBLoT substack newsletter.

You can also find my writing on Quora and Reddit.

How can I help you?

If you are an early stage start-up
You have a vision, some traction — maybe even a prototype — but you’re unsure how to implement machine learning effectively. I help you:

• Identify ML opportunities in your product or workflow (even if you're not sure where to start).

• Translate business goals into ML problem statements that can be solved within your current data and team constraints.

• Design and deploy MVPs of AI/ML systems that are lean, measurable, and scalable.

• Build foundational data pipelines, feature stores, or recommendation systems tailored to your product’s unique needs.

• Coach your founding or tech team on best practices around model evaluation, experimentation, and iteration — so you're not reliant on a black box.

📌 Example: At Blackboard Radio, I scaled an AI-powered communication coach that could assess and give real-time feedback on speaking fluency, all built in an agile product loop with the founding team.

If you are a mature company

As an enterprise, you likely already have data, dashboards, and maybe even some ML models. But real impact comes when these elements are deeply integrated into decision-making pipelines. I help you:

• Design and implement end-to-end ML pipelines — from data ingestion to model deployment and business reporting.

• Conduct AI opportunity audits across business units to identify and prioritize use cases with measurable ROI.

• Integrate ML solutions with your existing commercial, marketing, or ops systems, ensuring they’re not siloed.

• Improve model performance and adoption by aligning with business KPIs, not just technical metrics.

• Navigate stakeholder buy-in and change management, ensuring that business and tech teams are aligned.

📌 Example: At Pfizer, I’m working with their commercial analytics team to optimize revenue outcomes through a full-stack ML pipeline aligned with field team strategy and executive KPIs.

If you are a non-tech business wanting to explore AI/ML for your business  
AI doesn't need to be intimidating — or expensive. If you're a non-tech founder, SME, or traditional business curious about the buzz around AI, I help you:

• Understand where AI can realistically add value in your workflows (and where it won’t).

• Run pilot projects that validate ideas quickly with minimal investment.

• Evaluate vendors and tools to avoid overhyped or mismatched solutions.

• Build internal capacity by helping you hire or upskill the right talent.
  
  

📌 Example: With Mandrake Bio, I helped a non-technical founding team define and build an AI-powered scientific network graph that became a critical system to design new variety of drought resistant and high nutrition crops.

My Projects

• Scalable Bayesian Hyperparameter Optimization for Legacy ML Pipelines

  1. Integrated Bayesian hyperparameter tuning using Optuna into a large, complex codebase to accelerate model training workflows. 

  2. Navigated a deeply nested object-oriented architecture with intricate inheritance hierarchies, ensuring clean integration without disrupting core logic. 

  3. Designed modular wrappers to plug Optuna’s search space and optimization loop directly into existing training routines. 

  4. Achieved a 30% reduction in model iteration time, enabling faster experimentation with minimal refactoring. Enhanced reproducibility and scalability of the ML pipeline while maintaining codebase stability and team-wide usability.
     
     

• Deep Learning–Driven Subscriber Segmentation for Email Engagement Optimization

  1. Designed and deployed a newsletter engagement classifier using PyTorch to identify high-engagement subscribers, driving personalized email targeting strategies. 

  2. Engineered features from user interaction logs, content consumption patterns, and historical engagement data. 

  3. Modeled using a neural network architecture optimized for imbalanced classification. Integrated predictions into the marketing automation pipeline, enabling dynamic segmentation of the mailing list. 

  4. The model’s precision allowed for targeted campaigns, resulting in a significant increase in open rates—from 20% to 40%. Focused on feature engineering, model interpretability, and real-time inference to ensure marketing agility. 

  5. Deployed with monitoring hooks to track performance drift and retrain triggers. This project showcased the business impact of applied machine learning in user engagement and retention.
     
     

• Real-Time Audio Fluency Scoring Engine for English Language Learners

  1. Developed and deployed an automated audio-based fluency scoring systemto evaluate spoken English proficiency, replacing a manual review process that previously took up to 2 hours. 

  2. Leveraged signal processing, speech-to-text transcription, and linguistic feature extraction (e.g., speech rate, pause frequency, filler words) to quantify fluency. 

  3. Combined rule-based logic with lightweight ML models to score users on clarity, pace, and coherence in near real-time. 

  4. Integrated with a user-facing application for immediate feedback, enhancing the learning experience and user retention. 

  5. The solution was optimized for scalability and latency, ensuring consistent performance across varied network conditions and audio qualities. This system significantly improved operational efficiency, enabled continuous user progress tracking, and laid the groundwork for personalized speech improvement recommendations.
     
     

• Resolving author name disambiguation in PubMed

  1. Built a scalable ML pipeline for author name disambiguation in PubMed, addressing challenges like combinatorial explosion, semi-structured data, and internal dependencies (e.g., co-author networks). 

  2. Traditional classification failed due to non-categorical/non-numeric inputs and non-linear generalization. Developed a hybrid system using gradient boosting, graph clustering, and NLP, combined with a human-in-the-loop feedback loop to refine model predictions. Engineered an end-to-end automated pipeline with Docker, and AWS, designed for real-time scalability and production integration. 

  3. Achieved over 65% reduction in manual verification, enabling accurate author resolution for downstream applications like KOL discovery, expert search, and publication trend analysis in life sciences.
     
     

• Constraint based metabolic network analysis of Cancer cell lines 

  1. At Imperial College London, led computational research applying network science, graph theory, and linear programming to model and analyze cancer metabolism. 

  2. Simulated thousands of gene knockout scenarios across 100+ cancer cell lines using constraint-based modeling to identify therapeutic targets. 

  3. Integrated biological networks with flux balance analysis (FBA) to study system-level metabolic shifts. Published first-author peer-reviewed article in Frontiers in Bioengineering and Biotechnology (Impact Factor: 6.7), highlighting opportunities at the intersection of systems biology and computational optimization. 

  4. Gained deep expertise in biological data modeling, metabolic pathway simulation, and biomedical informatics, with strong emphasis on reproducibility, scientific rigor, and hypothesis-driven data science.

Academic Achievements

• Graduated with Distinction - Imperial College London

• Graduated with Honours - IIT Kharagpur

• Selected among 100 Leaders of Tomorrow - Global Biotech Revolution

• Kishore Vigyan Protsahan Yojana Awardeev

• Published a First Author article in Peer peer-reviewed journal.

• GRE score 330 - First attempt