Headline

Data Scientist / ML Engineer | RAG, LLM Fine-tuning, MLOps | Python, PyTorch, AWS/GCP/Databricks | BITS Pilani

About

I'm a Data Scientist with 4+ years building and productionizing ML/LLM systems at scale: spanning RAG pipelines, LLM fine-tuning, and MLOps across AWS, GCP, and Databricks. At CommerceIQ, I own the full lifecycle of our retrieval and classification systems: running RAG experiments across multiple LLMs and multimodal product data that lifted accuracy from 65% to 88%, cutting LLM inference latency by ~50% through an async pipeline redesign, and building the ingestion infrastructure that keeps our taxonomy's vector DB continuously fresh. Before this, at Swift, I built forecasting and classification models (route-choice + quantile regression for delivery time estimation, image classifiers for weight verification, risk models for return prediction) that moved real operational metrics — SLA hit-rate, revenue per shipment, RTO rates. I care about the full ML lifecycle, not just modeling: data and feature engineering, evaluation design, deployment, and the MLOps that keeps a model reliable in production. Comfortable owning a problem end-to-end, from a messy dataset to a shipped API. Currently exploring Data Scientist / ML Engineer / Applied Scientist roles where I can keep working across the LLM + production ML stack, and available to join immediately. Open to connectin. Reach me at tiwarimanas866@gmail.com.

Location

Bengaluru

AI stack

claude code

hugging face

ollama

Work experience

Data Scientist 2 at CommerceIQ, Apr 2026 - Present · 6 mos, Bengaluru · Hybrid

Data Scientist - 1 at CommerceIQ, Aug 2024 - May 2026 · 1 yr 10 mos, Bengaluru · Hybrid, - Ran RAG experiments on Databricks across multiple LLMs and multimodal SKU metadata (product description, bullet points, image), lifting accuracy 65%→88%, coverage 43%→62%, and high-confidence accuracy 70%→95%; cut LLM inference latency ~50% via an asynchronous pipeline redesign — a reduction that held even after upgrading to a larger model (Gemini 2.0 Flash → 2.5 Flash). - Productionized an Airflow/AWS/Databricks ingestion pipeline that continuously refreshes our product taxonomy's vector DB, replacing a static index with dynamic updates and a feedback loop from manually corrected labels — improving retrieval accuracy by 5-8 basis points. - Ran controlled fine-tuning experiments on a BERT-based classifier, achieving 88% accuracy across ~400 classes. - Constructed golden evaluation datasets via stratified sampling, replacing fully manual audits and cutting evaluation time from 1-2 weeks to a few hours. - Built an error-analysis module that shifted first-level debugging of market-share inaccuracies from Data Science to Operations, saving the DS team 2-3 days/week. - Built a weekly accuracy/coverage reporting module based on Operations team annotations.

Data Scientist at Swift, Jun 2023 - Jul 2024 · 1 yr 2 mos, Bengaluru, - Built a probabilistic ETD (estimated delivery time) model combining a route-choice model with segment-level quantile regression, achieving an 88% SLA hit-rate. - Implemented a Multi-Criteria Decision Making (MCDM) model for smart courier allocation, delivering a 20% improvement in seller platform performance. - Built an image classifier via transfer learning on EfficientNet (TensorFlow), achieving 93% accuracy — generating ~₹2 lakh/month in cost savings by resolving weight disputes. - Built an XGBoost-based weight-prediction classifier (92% accuracy), increasing revenue per shipment by ₹4.5. - Developed a Multi-Layer Perceptron model for RTO (Return to Origin) risk classification, achieving 83% accuracy / 74% precision and reducing seller RTO rate by 4%. - Optimized BigQuery queries, cutting monthly costs by 30%. - Managed a self-hosted Kafka cluster with automated AVRO schema generation, cutting connector setup time by 3-4 hours. - Built a jobserver orchestrating ETL and MLOps pipelines using async Python.

Business Analyst at Swift, Jul 2022 - Jun 2023 · 1 yr, Bengaluru, Karnataka, India

Business Analyst at Swift, Apr 2022 - Jun 2023 · 1 yr 3 mos, Bengaluru

Intern at DHIO Research & Engineering Pvt Ltd.,, May 2020 - Jul 2020 · 3 mos, Mumbai City district, Maharashtra, India

Education

Bachelor of Engineering, Mechanical Engineering, BITS Pilani, Hyderabad Campus

Public links

github: https://github.com/glitch-8