Akaike (A boutique data science consulting startup) | Bengaluru

Principal Data Scientist

• Established and led a 3-member consulting vertical, designing the end-to-end engagement framework and securing the first client win - a Pharma Next Best Action study, bringing in ~$250K revenue.
• Managed a 12-member cross-functional data science team (5 tiers, including indirect reportees), running regular retros for team health checks and monthly 1:1s for bidirectional feedback and junior talent development. Actively contributed to employee retention through counseling and coordination with leadership.
• Overhauled the DS hiring process across Standardization, Process, and Governance pillars, cutting hiring TAT from 8-10 weeks to 3-5 weeks, with unified evaluation frameworks, structured round formats etc. Revamped performance reviews from a two-way model to a structured 360-degree system - building a custom rating normalization model across Cultural, Technical, Growth, and Leadership dimensions - reducing review complaints and democratizing feedback beyond the reporting manager.
• Built and institutionalized an assetization framework, producing 4 standardized artifacts (codebase, whitepaper, executive deck, video demo) per project — successfully assetizing 10+ projects which got leveraged across Data Science, Sales, and Marketing teams.
• Pitched and delivered a Pharma Next Best Action system for sales rep-to-HCP engagement across Remote, In-Person, and Email channels while working as Engagement Manager cum Principal Data Scientist - architecting a 12-component prescriptive ML system (sequential-ensemble neural network) that drove $20M/disease area/annum/region impact across top 3 brands, validated through retrospective A/B testing.
• Led delivery of a Market Basket-based medical code extraction product as its Engagement Manager - replacing a 3 person, month-long manual process with an automated pipeline, achieving 80% OPEX reduction and ~90% team adoption.
• Led demand forecasting solution design for an Indian QSR startup - replacing an over/under-forecasting system to drive 20% topline growth and 45% reduction in perishable raw material wastage, managing end-to-end client engagement as both team lead and thought leader.
• Led HSI-based object detection for a client in Defense sector - directing model design and annotation for 4-5GB hyperspectral satellite imagery, coaching the team on specialized CV methodologies uncommon in standard workflows.

McKinsey & Co. (A multinational strategy and management consulting firm) | Gurugram

Sr. Analyst, Expert Consulting

• Owned end-to-end credit risk models development for two business lines - Secured & Unsecured for an Indian NBFC - integrating application, cashflow, bureau, and financial datasets into a modular PD model handling both NTC and ETC customers, delivering 40%+ GINI improvement over existing models and 85% reduction in application decisioning TAT.
• Built PySlice - an internal Credit Risk Modelling accelerator integrating data connectors, WoE binning, feature selection, and modelling pipelines - cutting first-cut model development from 3 months to 1 week, with an additional data synthesizer module (PGMs + distribution estimation) extending its utility across 10+ banking client engagements.
• Led Anode Tracking system design for an aluminum smelter across Dubai & Abu Dhabi (3-month deployment) - coordinating camera setup, data pipelines, and annotation, and building a multi-stage CV pipeline achieving 92% anode tracking rate from paste plant to baking furnace.
• Led Fleet Optimization for an Oil & Gas major in South America - formulating a LP-based allocation model (reduced from 10K+ to ~1K decision variables), with Monte Carlo sensitivity analysis, driving ~$30M OPEX reduction; also deployed to Malaysia for client capacity building and authored a whitepaper which was eventually used as a foundation for an internal firm asset.
• Led demand forecasting for an Oil & Gas major's polyethylene business - designing a spatio-temporal model that predicted end-product demand across hundreds of SKUs to optimize working capital, while managing SteerCo communications and leading a 2-member on-ground team within a broader 8-member engagement.
• Built a Total Commercial Spend Optimization (TCSO) prescriptive ML model for an Indian Lifesciences major — layering an optimization routine over an MMM model (AdStock, Hill Saturation, Bayesian tuning) across Linear Regression and GAM archetypes, with a post-engagement whitepaper repurposed as a firm-wide asset.

Roadzen (AI-driven platform in auto insurance and mobility space) | New Delhi

Data Scientist

• Built a custom OCR solution using CNN-based classification and localization - reducing inspector form-filling time and driving ~400% throughput increase in the treatment group vs. control, validated through A/B testing.
• Designed a human-mimicking KV-extraction framework for 200+ policy document layouts - building a config-driven, end-to-end PDF processing engine (readable and non-readable) capable of extracting key-value pairs in seconds per document.
• Built an internal DL classification accelerator - cutting first-cut model experimentation TAT by 85%.
• Led and streamlined the annotation pipeline, coordinating a 12-member annotation team for ground truth labelling.

Nissan Digital (Nissan's digital innovation center in India) | Trivandrum

AI Engineer, Data Science CoE

• Led demand forecasting for 9 car models across NNA and Japan — reducing MAPE from 34% to 8% and driving 8% inventory cost reduction (~$100M savings), transitioning from classical time-series models to XGBoost/ML methods with custom features for external shocks and marketing events.
• Led end-to-end deployment of a CV-based corrosion detection system on the factory floor - from camera setup and annotation to training deep learning models (CNN to complex architectures) - doubling inspection throughput and significantly improving quality.
• Built a kiosk-based car recommendation engine (Recall@10: 63%) - architecting a neural network for rank-ordered model recommendations with a custom de-clustering algorithm for granular outputs, integrated into the customer-facing kiosk in collaboration with UX/UI, Data Engineering, and Software Engineering teams.
• Led the development of a predictive maintenance system for factory motors - streaming accelerometer data via Raspberry Pi to cloud, building a density-based anomaly detection model with real-time mobile alerts, reducing overhaul incidences by ~40%.
• Represented Nissan's Data Science CoE at conferences and university lab visits, strengthening the talent pipeline and employer brand.

• Promoted from Fellow to Sr. Analyst within 1 year of joining McKinsey (Joining: Jan’22, Promoted in Dec’22).
• Gave a 5 hour conference talk on time series forecasting at GIDS, 2020, titled “Little Data, Huge Expectations”
• Created two open-source libraries:
  • swtLOC: A python Implementation of Stroke Width Transform algorithm from Microsoft Research, Detecting Text in Natural Scenes with Stroke Width Transform, 2010. 
  • tsEuler: A python Library for time series analysis dashboard.

• Bachelors in Technology (Electrical Engineering), Jamia Millia Islamia, 2018.

• Post Graduate Diploma in Applied Statistics, IGNOU, 2021
• Deep Learning Specialization, Deeplearning.ai, 2019
• Machine Learning, Stanford, 2019
• Mathematics for Machine Learning, Imperial College, 2019
• Qiskit Global Summer School on Quantum Machine Learning, IBM, 2021

• Successfully trekked to Everest Base Camp (5364 mt) in December’23 and Summitted Kala Patthar (5700 mt) in December’23.