Rishabh Thukral

Senior AI Engineer

Responsibility: Contributed to building Briink’s Agentic AI Platform for ESG analysts, enabling questionnaire pre-filling using advanced RAG and multi-step reasoning techniques. Improved system F1 score from 0.65 to 0.85+ for ESG use cases and co-led the rollout of an integrated evaluation platform and LLM observability, enhancing accuracy and transparency, supporting their seed round of 3.8Me.

Tech Stack: Python Stack(PyTorch, FastAPI etc), Docker, GCP(Cloud SQL, Cloud Run, AI studio etc), Modal, VectorDB (Elastic Search), LLM stack (OpenAI, Anthropic, Langchain, Langsmith, CrewAI, Cohere, Open source rerankers - BGE, MiniLM etc), Segment, June Analytics

Independent AI Consultant

Responsibility: Instrumental in developing company’s AI Analyst using SOTA LLMs, spearheaded data pipeline creation for ingesting and embedding data for establishing surplus. Established and improved RAG-based chat system and explored model fine-tuning. Additional applications involve response clustering, insight-driven research points, enhancing our SaaS platform’s user experience and efficiency. Contributed in scaling the product from zero to 45K$ MRR.

Tech Stack: Python (PyTorch, Pandas, FastAPI etc), Docker, AWS (S3, ECS, StepFunctions, ECR, Bedrock), Modal, Heroku, VectorDB (Chroma, Qdrant), OpenAI’s API Stack

Senior Data Engineer - AI Factory

Responsibility: Contributed to the integration of real-time data transformation and ingestion pipelines, devised alongside Data Scientists, to integrate advanced ML models (for anonymisation, trust score prediction on diagnoses, symptom search etc) into our product as part of our interoperability stack used by our clinical partners. Later, led the roll-out of an LLM RAG-based chat service for the medical sector, focusing on enhancing its performance with SOTA LLMs and other fine-tuned models.

Additionally, served as the Team Lead for AI Factory for the last 5 months, overseeing project management and team coordination.

Tech Stack: Python (Spark, PyTorch, Pandas, etc), Kotlin(Spring, HapiFHIRR, etc), Tableau, Docker, AWS (MSK (managed Kafka), EKS, Athena, StepFunctions, EMR, Bedrock etc), Terraform, ClearML, HL7 - FHIR Standard, Chroma VectorDB, OpenAI’s API Stack

Data Engineer

Responsibility: Spearheaded the deployment of version 2 of our analytics infrastructure, involving with collaboration of other senior engineers. Conceptualized, implemented and maintained the data infrastructure for business needs. Collaborated closely with data scientists and analysts to develop state-of-the-art machine-learning systems for demand forecasting and case refinement predictions. Was part of the company until it got acquired in a 138Me deal by Straumann.

Tech Stack: Python, Airflow, SingerIO, DBT, Looker, Metabase, Airbytes, Docker, AWS (EC2, S3, ECS, Redshift, DMS, SageMaker), GCP(BigQuery, Analytics platform) Gitlab, CI/CD, Matillion

Data Engineer

Responsibility: Constructed diverse Extract, Transform, Load (ETL) pipelines enhancing our proprietary ”Data Engine” deployed on a self-hosted Airflow cluster. Additionally, we addressed the critical need for temporal tracking of company features, solidifying our data infrastructure targeted at this.

Tech Stack: Python, Airflow, Pandas, NumPy, BeautifulSoup, Django, RecordLinkage, SciKit Learn, Custom Search Engines, PostgreSQL, DynamoDB, WARC Archiving, AWS (EC2, S3, Eks, RDS, Redshift), Sentry

Data Engineer

Responsibility: Contributed to the creation of a multi-tier proprietary data engine enabling API integration for early-stage startup data collection. Spearheaded internal API/webhook integration and assisted in ML research for investment rankings. Our strategic development of the DA engine, Sofia, catalysed the company’s evolution from a VCaaS with zero to 500Me AUM with multiple sector funds.

Tech Stack: Python, Airflow, Pandas, NumPy, Beautiful Soup, Flask, Sci-Kit Learn, MySql, MongoDb, HDFStores, AWS (EC2, S3, RDS, SNS), SPIKEs with Neo4J, GraphQL, CommonCrawl WARC formats.

Reviewer and Mentor

Student Mentor: Mentored and reviewed projects of hundreds of students enrolled in Udacity’s Intro to Self Driving Car ND, MLND & AI programming with python ND.

Tech Stack: Python, Pandas, NumPy Beautiful Soup, Flask, Sci-Kit Learn, PyTorch, TensorFlow, TensorBoard, HDFStores, AWS (EC2 (GPU Computes), S3, RDS, SNS), Neural Networks - Deep CNN architectures, DQNs, Skip-Layer architectures, Fully Convolutional Architectures for Semantic Segmentation

Student Research Assistant

Detecting lightning strikes in high-speed videos: Built an end-to-end classifier using Deep CNN Architectures to identify given any image is a lightning image or not. This model provides the basis for the smart data warehousing for the project.

Tech Stack: Python, Pandas, NumPy, Flask, PyTorch, TensorBoard, MySql, SQL Alchemy, OnPrem data centres for remote computations, Neural Networks - Deep CNN architectures, Skip-Layer architectures (For image classification purposes)

Backend - Systems Engineer

Feature Extraction and Detection: from the text in the context of recorded food deals from various websites to analyze the nature and information about it from the data. Developed a Facebook Messenger bot using basic NLP and Decision Flow implementation. The bot allowed any Facebook user to get food media updates from the Ketchupp Blog and access many other exciting features.

Tech Stack: Python, Pandas, Beautiful Soup, Flask, SQL Alchemy, NLTK, AWS (EC2, S3, RDS, SNS), Facebook Messenger Platform, REST APIs

CGC - College Of Engineering

Built an end to end classifier using pretrained Deep CNN Architectures to identify whether any given frame from the video is a lightning image or not. Also working on the analysis of corresponding Electric - Field data using LSTM architecture in order to make the current results better.

Developed and trained a Feed Forward Classifier Network on top of VGGNet to label the given image of a flower as one of the 102 categories available in the 102 Category Flower Dataset. Trained the model to get an accuracy of ~72%. Currently working on getting state of the art results using ResNet.

Developed the project following the description of the Deep Q Learning algorithm (Deep Mind) described in the Playing Atari games with Deep Reinforcement Learning. This project shows that this learning algorithm can further generalized to the notorious Flappy Bird.

Developed a Fully Convolution Network on top of VGGNet using skip layer architecture to label the pixels of a road in a image. This project can be considered as an extension of Jonathan Long and Evan Shelhamer's work in the paper "Fully Convolution Network for Semantic Segmentation". The model was trained on Kitti road data set on an AWS GPU Machine.

Developed an android application which allows a user to learn 5 new words of English language every day through interactive activities like quiz based games and activities like fill in the blanks and match the following.

Feature Extraction and Detection from text data

Feature extraction and detection from the text in the context of recorded food deals from various websites for the purpose of analyzing the nature and information about it from the data. Further, the pipeline was connected to an EC2 instance on the cloud with seamless integration to the existing product.