Snigdha Sharma, Developer in Bengaluru, Karnataka, India
Snigdha is available for hire
Hire Snigdha

Snigdha Sharma

Verified Expert  in Engineering

Artificial Intelligence Developer

Location
Bengaluru, Karnataka, India
Toptal Member Since
August 23, 2023

Snigdha is an experienced engineer specializing in software development, NLP, computer vision, machine learning, and AWS technologies. She has worked with companies of many shapes and sizes, from a month-old startup still iterating on its ideas to a large bank like JPMorgan Chase to one of the fastest growing names in HR tech—Eightfold.ai. Snigdha has gained insightful experience in several industries, including pharmaceuticals, media and entertainment, national security, fintech, and HR tech.

Portfolio

Consulting
Amazon Web Services (AWS), Python 3, Generative AI, OpenAI, Machine Learning...
Eightfold AI
Python 3, OpenAI GPT-4 API, XGBoost, Spark, Software Development...
JPMorgan Chase
Amazon Web Services (AWS), Natural Language Processing (NLP), PyTorch...

Experience

Availability

Full-time

Preferred Environment

Linux, Git, Windows, Python

The most amazing...

...project I've led is a feature suite developed by leveraging ML and LLMs to automate the recruiting process for a high-growth HR-tech startup.

Work Experience

AI consultant | Fractional AI Head

2024 - PRESENT
Consulting
  • Developed a data analyst tool using generative AI that can talk to data while ensuring the explainability of the results and minimum hallucination.
  • Reformated unstructured documents to create clean, unambiguous, nonrepetitive, and well-structured documents.
  • Used RAG to develop a customer support bot that answers user queries, taking help from massive product documentation—with close to 0 hallucinations. Implemented testing pipelines to streamline QA for the chatbot.
  • Improved prompts for several clients using best practices and extensive insights gained by working on several prompting frameworks.
  • Successfully implemented several use cases in production using prompt engineering and generative AI including summarisation, qna, text evaluation, information extraction, code generation, data analysis, reasoning, personalisation and many more.
Technologies: Amazon Web Services (AWS), Python 3, Generative AI, OpenAI, Machine Learning, Embeddings from Language Models (ELMo), Retrieval-augmented Generation (RAG), Large Language Models (LLMs), AI Model Training, Prompt Engineering, Pinecone, Product Design, Product Ownership, Scalable Vector Databases, Multimodal Models, LangChain, Machine Learning Operations (MLOps), Distributed Systems, Pandas, SQL

Senior ML Engineer

2021 - 2024
Eightfold AI
  • Led a generative AI team to develop features that can transform how users interact with our product.
  • Implemented many use cases with prompt engineering and generative AI Thoroughly tested these prompt on production test datasets and achieved significant improvements as compared to generic prompt libraries out there.
  • Launched the product's first AI text-generation use case using OpenAI models and prompt engineering. Also implemented streaming to reduce effective latencies and caching to reduce OpenAI request costs.
  • Won the "Most Innovative Hack" award among 80 teams and led a team of two to re-imagine dashboards showing hiring analytics. I talked to stakeholders, understood pain points, and demoed an ML-based solution that has since been launched for our users.
  • Resolved customer issues, critical technical bugs, and on-call issues in both resume parsing and matching and ranking modules in specific and integration problems with other services in the product.
  • Implemented human-annotated datasets to track ML model launches, reducing harmful recommendations by approximately 2% and improving feature engineering.
  • Guided initiatives on improving job recommendations for internal employees and skill extraction improvements for better skill-based hiring.
  • Owned and delivered personal contact details and education extraction modules using entity extraction techniques, achieving a better AUC than third-party services. I also set up pipelines to track metrics and latencies in production.
  • Participated in the ML hiring pipeline and interviewed candidates, onboarding and mentoring interns and new graduates amongst the top talent pool.
  • Researched and presented learning sessions to the team to understand new ML solutions to product problems.
Technologies: Python 3, OpenAI GPT-4 API, XGBoost, Spark, Software Development, Machine Learning, Software Design, Linux, Git, Entrepreneurship, Scikit-learn, SpaCy, Natural Language Processing (NLP), Amazon Web Services (AWS), Named-entity Recognition (NER), Indexing, Tf-idf, DistilBERT, CI/CD Pipelines, Generative Pre-trained Transformers (GPT), OpenAI GPT-3 API, Team Leadership, Planning, Stakeholder Management, ChatGPT, Artificial Intelligence (AI), Python, Generative Pre-trained Transformer 3 (GPT-3), Audio, OpenAI API, Data Analysis, Feature Engineering, MacOS, Flask, Deep Learning, Computer Science, Regression, Hypothesis Testing, Data Science, Data Analytics, Data Visualization, Statistics, Chatbots, Chatbot Conversation Design, Data Engineering, ETL, Cloud Architecture, Recommendation Systems, Artificial General Intelligence (AGI), APIs, Technical Leadership, Leadership, Mentorship & Coaching, Large Language Models (LLMs), OpenAI, Software Architecture, Cloud, Algorithms, Models, Generative AI, Generative Artificial Intelligence (GenAI), LoRa, Hugging Face, Retrieval-augmented Generation (RAG), BERT, AI Model Training, Prompt Engineering, Pinecone, Product Design, Product Ownership, Scalable Vector Databases, LangChain, Machine Learning Operations (MLOps), Distributed Systems, Pandas, Docker, SQL

Data Scientist

2020 - 2021
JPMorgan Chase
  • Envisioned and developed a product to derive insights from earning call transcripts for around 500 public companies using prompt engineering. I extensively researched third-party tools to create product features and architecture design documents.
  • Introduced a pipeline for information retrieval from PDF documents using GPT-2, DistilBERT, and text similarity using BERT embeddings. Analysts can use the system to create reports, saving almost 70% of manual reading effort.
  • Leveraged a graph variational auto-encoder to calculate node similarity in a knowledge network graph of potential clients.
  • Launched a NER API using Python, TensorFlow, and MLflow to automatically annotate documents using some pre-defined labels and saving hours of annotation effort.
  • Deployed multiple apps using CI/CD in Jenkins on the internal Kubernetes platform. I devised interactive dashboards to analyze client emails in shared mailboxes and present insights from earning call transcripts for around 500 public companies.
  • Held the top position with teams in internal hackathons.
Technologies: Amazon Web Services (AWS), Natural Language Processing (NLP), PyTorch, TensorFlow, SpaCy, Named-entity Recognition (NER), Sentiment Analysis, Indexing, Tf-idf, DistilBERT, MLflow, Graph Neural Networks, CI/CD Pipelines, Kubernetes, Software Development, Machine Learning, Software Design, Linux, Git, Windows, Python 3, XGBoost, Scikit-learn, Generative Pre-trained Transformers (GPT), Planning, Stakeholder Management, Artificial Intelligence (AI), Python, Generative Pre-trained Transformer 3 (GPT-3), OpenAI API, Data Analysis, Feature Engineering, Flask, Deep Learning, Computer Science, Regression, Hypothesis Testing, Data Science, Data Analytics, Data Visualization, Statistics, Data Engineering, ETL, Cloud Architecture, Recommendation Systems, Artificial General Intelligence (AGI), TensorFlow Deep Learning Library (TFLearn), APIs, Technical Leadership, Leadership, Finance, Mentorship & Coaching, Large Language Models (LLMs), OpenAI, Software Architecture, Cloud, Algorithms, Models, Generative AI, Generative Artificial Intelligence (GenAI), Hugging Face, Retrieval-augmented Generation (RAG), BERT, Text to Speech (TTS), AI Model Training, Prompt Engineering, Product Design, Product Ownership, Machine Learning Operations (MLOps), Distributed Systems, Pandas, AWS Lambda, Docker, SQL

Senior Product Engineer

2019 - 2020
Myelin Foundry
  • Researched current architectures and performed E2E ML modeling to train content-aware neural network-based super-resolution models that can be applied to videos in real time to improve the OTT media user's experience and reduce bandwidth usage.
  • Released a fully functional POC demonstrating the above capability for desktop and Android, which earned the project its first client, a top video streaming company.
  • Delivered a POC to de-noise SAR satellite images using U-Net-like architecture.
Technologies: Computer Vision, Natural Language Processing (NLP), Python 3, TensorFlow, Software Development, Machine Learning, Software Design, Linux, Git, Entrepreneurship, Azure, PyTorch, Planning, Stakeholder Management, Artificial Intelligence (AI), Python, Data Analysis, Feature Engineering, Flask, Convolutional Neural Networks (CNN), Deep Learning, Computer Science, Data Visualization, Image Processing, Artificial General Intelligence (AGI), Generative Adversarial Networks (GANs), APIs, Technical Leadership, Leadership, Mentorship & Coaching, Large Language Models (LLMs), Software Architecture, Cloud, Algorithms, Models, Generative Artificial Intelligence (GenAI), BERT, Text to Speech (TTS), AI Model Training, Product Design, Product Ownership, Machine Learning Operations (MLOps), Pandas, Docker

Associate Software Developer – Research and Development

2017 - 2019
Axtria
  • Prepared a business case study for sales prediction based on investments in different promotional channels and lagging sales.
  • Trained a custom neural network in TensorFlow and visualized results in TensorBoard with a 71% reduction in error.
  • Developed a modular, meta-data driven, multi-threaded, scalable Flask app for business rule execution to automate sales crediting and incentive compensation processes, integrating Salesforce and Amazon S3, Redshift, and Athena.
  • Designed a wrapper web app for Amazon Polly, introducing additional features such as parsing text from single or multiple files to convert it into downloadable MP3 speech files for internal usage.
Technologies: Amazon Web Services (AWS), Python 3, Redshift, PostgreSQL, Salesforce, Software Development, Machine Learning, Software Design, Git, Windows, Entrepreneurship, XGBoost, Scikit-learn, TensorFlow, Planning, Stakeholder Management, Artificial Intelligence (AI), Python, Audio, Data Analysis, Feature Engineering, Flask, Deep Learning, Computer Science, Regression, Data Science, Data Analytics, Data Visualization, Statistics, Data Engineering, ETL, Cloud Architecture, TensorFlow Deep Learning Library (TFLearn), APIs, Mentorship & Coaching, Software Architecture, Cloud, Algorithms, Models, Text to Speech (TTS), AI Model Training, Pandas, AWS Lambda, SQL

GenAI Copilot Suite for HR-tech Product

This project involved leading a team of four members to power an HR-tech product with generative artificial intelligence (GenAI) capabilities and make recruiting easier for our enterprise customers.

I developed the back-end software design, engineered LLM prompts, discussed improvements and feasibility with our UX design team, got my team unblocked on development issues (both back-end and front-end development), mentored interns to develop utility frameworks and modules, conducted PR reviews, discussed features to be developed with stakeholders.

The developed functionalities include:
• Creating text from scratch for custom use cases.
• Inferring user intent and refining existing templates for personalization and customization in just a few seconds.
• Summarizing information required to make hiring decisions all in one place.
• Making inferences and answering questions from long text.

All these capabilities are offered while ensuring the entire system is fit for enterprise use cases, removing confidential information before sending it to LLMs and maintaining the professionalism and accuracy of the outputs generated to be scaled for business use cases.

Industry Insights Using Earning Call Transcripts

An in-house product that I envisioned and developed using natural language processing algorithms. The product aims to derive insights from earning call transcripts for around 500 public companies.

I worked independently on this project, communicating with the senior leadership at JPMC. Also, I conducted extensive research on the 3rd-party tools available to create Product Feature Documents and Architecture Design documents as future work to convert the above project into a competing product at scale.

Named Entity Recognition

Independently launched a named entity recognition (NER) API using Python, TensorFlow, and MLFlow. The API:
• Automatically annotates documents by extracting entities from a database, saving approximately 500 hours of manual annotation work.
• Trains and fine-tunes NER models using Bi-LSTM CNN.
• Stores, manages, and facilitates model version comparisons.

This tool was applied to train models for extracting product opportunities from meeting notes, call transcripts, and automatically tagging client emails for automatic ticket creation.

Fovea | Video-resolution Enhancement Tool

https://www.myelinfoundry.com/solutions/fovea-stream/
This project involved creating an Android app—later converted into a plugin for some existing streaming apps—that transforms a low-resolution video to a higher resolution in real time. I mainly leveraged CNN-based neural networks run on Snapdragon.

I was a senior developer on a team project with 3-4 members, focusing on researching various CNN-based super-resolution architectures. We trained a CNN-based model best fit for our use case, maximizing quality while keeping the inference time low enough to allow real time video super resolution at 30fps. I integrated this model with a video and image library. Also, I integrated this in a dummy Android app to demonstrate the feasibility/quality of this model when run on Android devices. Running this on Android devices posed some challenges wherein we had to utilize a DSP runtime in Snapdragon processors, for which our team worked with some industry experts.

Additionally, I conducted research on speeding up the process by leveraging knowledge of video encoding, including i/p/b frames.

Matching Candidates with Positions

I played a key role in a project that ranks open positions for candidates and vice versa to identify the best matches. My responsibilities included substantial enhancements to the feature engineering pipelines powering our ML models.

During this project, I:
• Spearheaded skill extraction improvements to facilitate more effective skill-based hiring.
• Led efforts to enhance job recommendations for internal employees.
• Launched human-annotated datasets, specifically tailored for internal job postings, ensuring they accurately represent real-world data. Also, I developed an objective evaluation system that the annotation team used to create a high-quality evaluation dataset. Additionally, I created a user-friendly Flask service to streamline the annotation process.
• Successfully reduced bad job recommendations for internal employees by approximately 2% through feature engineering improvements.
2013 - 2017

Bachelor's Degree in Computer Science

National Institute of Technology, Delhi - New Delhi, India

DECEMBER 2023 - PRESENT

Brand and Product Management

IE Business School

DECEMBER 2022 - PRESENT

Entrepreneurship Specialization

Wharton University of Pennsylvania | via Coursera

MAY 2021 - PRESENT

Practical Reinforcement Learning

Higher School of Economics

JANUARY 2021 - PRESENT

Bayesian Methods for Machine Learning

Higher School of Economics

DECEMBER 2020 - PRESENT

Program on Business Analytics and Machine Learning for Asset and Wealth Management

Indian Institute of Management Bangalore

DECEMBER 2020 - PRESENT

How to Win a Data Science Competition

Higher School of Economics

MAY 2020 - PRESENT

Introduction to Deep Learning

Higher School of Economics

DECEMBER 2019 - PRESENT

General Course on Intellectual Property

WIPO Academy

Libraries/APIs

XGBoost, Scikit-learn, SpaCy, PyTorch, TensorFlow, Pandas, TensorFlow Deep Learning Library (TFLearn), React

Tools

Git, Named-entity Recognition (NER), ChatGPT

Frameworks

Flask, Spark

Languages

Python 3, Python, SQL, R

Paradigms

Data Science, ETL

Platforms

Linux, Windows, Amazon Web Services (AWS), MacOS, AWS Lambda, Docker, Azure, Kubernetes, Salesforce

Storage

Redshift, PostgreSQL

Other

Software Development, Machine Learning, Software Design, OpenAI GPT-4 API, Natural Language Processing (NLP), Indexing, DistilBERT, MLflow, OpenAI GPT-3 API, Generative Pre-trained Transformers (GPT), Planning, Stakeholder Management, Artificial Intelligence (AI), Generative Pre-trained Transformer 3 (GPT-3), OpenAI API, Data Analysis, Feature Engineering, Convolutional Neural Networks (CNN), Deep Learning, Computer Science, Regression, Hypothesis Testing, Data Analytics, Data Visualization, Statistics, Chatbots, Chatbot Conversation Design, Image Processing, Data Engineering, Cloud Architecture, Recommendation Systems, Artificial General Intelligence (AGI), APIs, Technical Leadership, Leadership, Mentorship & Coaching, Large Language Models (LLMs), OpenAI, Software Architecture, Cloud, Algorithms, Models, Generative AI, Generative Artificial Intelligence (GenAI), LoRa, Hugging Face, Retrieval-augmented Generation (RAG), BERT, Text to Speech (TTS), Prompt Engineering, AI Model Training, Product Design, Product Ownership, Multimodal Models, LangChain, Machine Learning Operations (MLOps), Entrepreneurship, Computer Vision, Graph Neural Networks, CI/CD Pipelines, Team Leadership, Audio, Generative Adversarial Networks (GANs), Finance, Pinecone, Scalable Vector Databases, Distributed Systems, Sentiment Analysis, Tf-idf, Reinforcement Learning, Deep Reinforcement Learning, Bayesian Statistics, Bayesian Inference & Modeling, Intellectual Property, Brand Management, Embeddings from Language Models (ELMo)

Collaboration That Works

How to Work with Toptal

Toptal matches you directly with global industry experts from our network in hours—not weeks or months.

1

Share your needs

Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2

Choose your talent

Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3

Start your risk-free talent trial

Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.

Top talent is in high demand.

Start hiring