Experience
[WORK_EXPERIENCE_HISTORY]
Factity- NCL | Milton Keynes, UK | SEP 24 - Present
RESPONSIBILITIES
- Project 1: Developing a flood monitoring system for CMCL which includes real time analytics and geo json based interactive map.
- Project 2: The project involved developing a trading strategy using reinforcement learning to balance the trade execution vs the price short. I was responsible for developing documentation, algorithm and the assessing the metrics against twap and vwap strategies.
- Project 3: Developing a retrieval augmented generation system and combining it with an ocr type system to bring text into preferred formatting.
- Project 4: Developing and deploying a machine learning based model that takes different environmental metrics that are received on the solar panels to predict the amount of energy that is set to transmit inside the storage unit.
- Project 5: Developing an ocr based system that allows the data to be extracted into a preset data format and retrieve the records from the raw data.
- Project 6: Tasked to develop a tool that allows the user to see different model capabilities against different bench marks across several years.
- Project 7: Developing a repository of questions and data collection strategy that allows users/ researchers to fill out questions about different ai safety domains in a gamified environment.
- Project 8: Evangelising about different ai safety domains, ai agentic capabilities, interpretability, fine tuning methods, tokenization etc and create developer engagement at conferences, Tasked with writing research proposals for grants, bids relating to software development projects etc.
ACHIEVEMENTS
Project 1: Developed a Flood monitoring system with real time responsive metrics that was able to alert flood stations with immediate warnings.
→ Tech stack: Python, Render redis, render plsql, persistent disks, git (CI/CD), cloud
Project 2: Developed trading strategies for optimal trade execution beating twap and vwap strategies using ai agentic systems for better execution strategy.
→ Tech stack: doc, python, modules - numpy, pandas, matplotlib, plotly, jinja, flask, git, redis , cloud
Project 3: Developed a rag based system with the ability to extract text understand the content and generated tailored responses according to the inserted documents.
→ Tech stack: langchain, ollama, sagemaker, python, S3, google storage, django, web frameworks (HTML, CSS, Javascript), hugging face models, transformer
Project 4: Developed a supervised learning model for predicting PV output from different environmental variables.
→ Tech stack: MLflow, sagemaker, scikit learn, python, redis, flask, front end frame works
Project 5: Developed an ocr system that helps to extract text from excessively blurry images and fills it in the appropriate text fields.
→ Tech stack: langchain, ollama, sagemaker, python, S3, google storage, django, web frameworks (HTML, CSS, Javascript), hugging face models, transformer, Reckognition
Project 6: Developed a comprehensive platform that analyzes AI progress across 2,827 records, 40 benchmarks, and 33 distinct capabilities. This dashboard consolidates real data from leading AI labs and monitor progress metrics.
→ Tech stack: MLflow, sagemaker, scikit learn, python, redis, flask, front end frame works, plotly, dash
Project 7: Developed a comprehensive data set of key research ai safety questions in areas like Security and defense, model understanding, Oversight and control, Risk assessment etc and created repositories for open source development.
→ Tech stack: Render, aws, NLP libraries, python, nextjs, mongodb, sentry etc
Project 8: Developed blogs and documentation websites for prompts, model tuning methods, interpretability, lora based methods, ai agents and developed token visualizers etc.
→ Tech stack: Nextjs, React, web frameworks, transformers, python, aws, llm apis
Figg Wealth | Birmingham, UK | OCT 23 - SEP 24
RESPONSIBILITIES
- Responsible for developing a comprehensive NLP - based dashboard utilizing local AI models (llama stack) to process news text from APIs like fin news, newsapi, yahoo finance, for real time sentiment analysis of FTSE 100 companies. Integrating time streamed news into a single process pipeline.
- Responsible for creating a ETL pipeline that takes in different aspects of the market movements and creating an algorithm that generates the polarity score for the particular FTSE 100 company.
- Responsible for developing a real time interactive interface for Portfolio managers to guage insight into the trading day's stock trends.
- Managed stakeholder relationships ensuring alignment with goals and responsibilities.
ACHIEVEMENTS
The news dashboard was able to showcase the latest news insights relating to the company sentiment associated with the particular time period.
→ Tech stack: REST frame work, python, fin news, yfinance, plotly, HTML, CSS, Javascript, Redis, lstorage
The automated text extraction and summarization pipelines were able to generate concise insights on market- moving events, enabling identification of polarity change using the semantic score calculator.
→ Tech stack: pandas, SQLAlchemy, plus Airflow providers for connectors (e.g., HTTP, S3, databases).
The interactive UI with plotly, dash featuring different sectors, time frames, sentiment thresholds to facilitate intuitive data exploration.
→ Tech stack: plotly HTML, CSS, Javascript, plotly backend, dash, python
Developed and deployed a machine learning driven dashboard to cluster FTSE 100 companies based on risk profiles.
Engineered a data pipeline to ingest and preprocess financial data, integrating key metrics and enhancing risk assessment accuracy with Yahoo Finance API.
Developed interactive dashboard with advanced visualizations, enabling stakeholders to explore and compare risk profiles dynamically, facilitating data-driven investment and risk-management strategies.
Blockchain Advisors - Steam house (think tank) | Birmingham, UK | FEB 24 - APR 24
- Automated scripts for document generation using docx modules (RPA task handling).
- Created investment reports for 100+ potential investment firms across EU & UK.
- Developed data cleaning pipelines using ETL principles leading to 50% reduction in data cleaning time.
- Tested and prototyped locally run LLMS using ollama and langchain.
Birla Institute of Technology | India | Jul 2021- Jul 2023
- Developed a physics-informed machine learning model for simulating population dynamics.
- Fine-tuning of parameters using nonlinear optimization techniques.
- Model development, boundedness of the solution, stability of the solution, sensitivity analysis with state- of-the-art modeling techniques.
- MCDM models for ranking regions based on carbon footprint.
[SOCIETY_WORK_LOGS]