Work Projects involved in

Team Collaboration

Intelligence Mapping Agent

Company: Potentiam Ltd

Timeline: Jan 2026 - Feb 2026

Work done:

  • Built a two‑stage embedding pipeline: internal catalog generation from JSON sources and per‑session market embeddings with persistence for replay and recovery.
  • Implemented multi‑format Excel ingestion with provider‑specific parsing, standard column mapping, and auto‑detection via header patterns and content analysis.
  • Designed semantic matching using cosine similarity with LLM reranking for low‑confidence matches, producing mapping outputs with confidence scores and reasoning.
  • Added persistent session embeddings and resume flow to continue matching after failures without re‑embedding.
  • Built a Flask web workflow with SSE progress streaming, background processing threads, and secure download routes.
  • Integrated Microsoft OAuth 2.0 (MSAL) for authentication with session‑based access control on processing routes.
  • Implemented structured logging with rotation and batch‑to‑single fallback embedding strategy for API resilience.

Challenges faced & how they were overcome:

  • Inconsistent market file formats → Column auto‑detection using header patterns and content analysis.
  • API batch failures → Per‑item fallback embedding with NaN‑safe guards for missing descriptions.
  • Long run times with failure risk → Persistent session embeddings and resume script to avoid rework.

Skills & Technology:

  • Languages & Data – Python, pandas, NumPy, SciPy (cosine distance).
  • Embeddings & LLMs – Google Gemini (text-embedding-004), rerank prompts.
  • Web & Auth – Flask (SSE streaming, background threads), MSAL (Microsoft OAuth).
  • DevOps – Gunicorn, Nginx reverse proxy, systemd, SSL, structured logging with rotation.

Org Chart Editor

Company: Potentiam Ltd

Timeline: Nov 2025 - Dec 2025

Work done:

  • Built a two‑tier system with Vite + React frontend and FastAPI backend, implementing a three‑step workflow: LinkedIn scrape, Excel/CSV enrichment, and org chart editing with export. Built as an upgrade to it's previous version.
  • Implemented LinkedIn scraping via Playwright automation with human‑like pacing, multi‑selector fallback, and auto‑deleted temporary XLSX exports.
  • Developed an LLM‑powered Excel formatter pipeline (Gemini with Ollama fallback) for title refinement, department mapping, reporting role inference, and leadership flagging with batch requests and in‑memory caching.
  • Created client‑side transformation to convert spreadsheet rows into d3-org-chart graph models with leader detection, department node creation, and fuzzy manager matching.
  • Built an interactive org chart editor with click‑to‑edit properties, cycle‑safe "Reports to" dropdowns, node addition/deletion controls, and inline data modification.
  • Added presentation export tooling: PNG export with aspect ratio cropping, SmartArt outline clipboard copy, and auto‑generated PPTX outlines.
  • Integrated Microsoft Azure AD (MSAL) authentication with guarded routes and implemented CORS, .env blocking, and temp file lifecycle management.

Challenges faced & how they were addressed:

  • Unreliable LLM JSON responses → JSON parser that strips markdown fences with safe fallback and Gemini‑to‑Ollama failover.
  • Ambiguous reporting lines → Fuzzy title matching with second‑pass LLM disambiguation for manager resolution.
  • Brittle LinkedIn scraping → Multiple XPath selectors, human‑like delays, retry logic, and optional headless mode.
  • Large chart editability → Grouped non‑leadership roles with PNG and SmartArt export paths.

Skills & Technology:

  • Frontend – React 19, Vite, React Router, d3-org-chart, MSAL (Azure AD).
  • Backend – FastAPI, Pydantic, Uvicorn, Pandas, openpyxl.
  • Automation & LLM – Playwright (Chromium), Gemini API, Ollama fallback, batch caching.
  • Export & DevOps – PNG export, pptxgenjs, CORS, file‑based logging, temp file management.

Individual Contribution

Automated Data Pipeline IV (RecruitCRM)

Company: Potentiam Ltd

Timeline: Jan 2026 - Feb 2026

Work done:

  • Built a modular Python ETL pipeline with separated layers for configuration, API access, transformation, database operations, and orchestration to extract RecruitCRM data into PostgreSQL.
  • Created a resilient API client with header‑based rate limiting, retry logic, pagination support, and content‑type validation for non‑JSON responses.
  • Designed a DataTransformer layer that flattens nested JSON, normalizes column names, infers types dynamically, and adds extraction metadata columns.
  • Built a dynamic schema evolution system with automatic column addition, batch upserts with deduplication, and connection pooling.
  • Added derived dimension tables (dim_jobs_pipeline_stages, dim_hiring_pipeline_stages, candidate_job_hiring_stages) for hiring analytics.
  • Implemented incremental streaming extraction with ETL state tables, candidate selection logic, and chunked processing for memory efficiency.
  • Added CLI flags for targeted "hiring stages only" runs and detailed logging with rotation and error‑only logs.

Challenges faced & how they were overcome:

  • Non‑JSON API responses → Content‑type checks with graceful error handling.
  • Schema inconsistencies → Flexible parsing logic for varying field names (status_id, data, stages).
  • Upsert collisions → In‑memory deduplication by primary key before batch inserts.
  • Dimension key mismatch → Schema aligned with synthetic keys for reliable joining.

Skills & Technology:

  • Languages & Data – Python, SQL, PostgreSQL (dynamic schema evolution, connection pooling, upsert logic).
  • APIs & ETL – RecruitCRM REST API (pagination, rate limiting), streaming ingestion, incremental sync.
  • Data Modeling – Fact/dimension modeling, synthetic key design, derived datasets.
  • DevOps – Shell runner, env‑based config, virtualenv, structured logging with rotation.

Automated Data Pipeline III (Docusign)

Company: Potentiam Ltd

Timeline: July 2025 - Aug 2025

Work done:

  • Created a Docusign Developer account and configured the application to securely authenticate using JWT-based OAuth 2.0, following Docusign’s latest REST API guidelines.
  • Successfully completed 20 required API test calls in the sandbox environment to enable production access.
  • Established a secure integration between Python scripts and the production Docusign API to extract envelope data (digitally signed employee contracts).
  • Automated the secure download and local storage of all envelopes, eliminating the need for manual downloads and significantly reducing operational overhead.

Skills & Technology:

  • Languages – Python.
  • APIs – Docusign API.
  • Workflow Orchestration – Scripted data extraction, auto-download, storage, error handling, logging, alerting.
  • DevOps & Scheduling – Ubuntu server, Cron jobs.
  • Security Considerations – Used Docusign’s recommended JWT OAuth 2.0 flow for secure token-based authentication.

Automated ELT Data Pipeline II (Beauhurst to HubSpot CRM)

Company: Potentiam Ltd

Timeline: June 2025 - July 2025

Work done:

  • Built a Python-driven ELT pipeline to extract companies and contacts data from Beauhurst API, storing raw and transformed data with advanced filtering, staging, and audit capabilities in PostgreSQL which is then updated into HubSpot CRM in near real-time.
  • Applied complex custom filtering logic to identify and select only the most relevant records for HubSpot updates, handling edge cases and exceptions in business rules.
  • Used the HubSpot API to programmatically create or update CRM records with high accuracy.
  • Implemented end-to-end workflow logging, error handling, and alerting to ensure data integrity and traceability.
  • Designed staging tables separate from audit tables, following best practices for data integration—allowing safe, repeatable syncs and full change auditability.
  • Delivered timely, accurate CRM updates, saving the company time and cost by automating manual processes and improving data reliability.

Skills & Technology:

  • Languages – Python, SQL.
  • Datastore – PostgreSQL (schema design, indexing, query optimization, staging/audit table design).
  • APIs – Beauhurst API, HubSpot API.
  • Workflow Orchestration – Scripted ETL orchestration, error handling, logging, alerting.
  • DevOps & Scheduling – Ubuntu server, Cron jobs.

Automated ETL Data Pipeline I (HubSpot CRM to PostgreSQL & Power BI)

Company: Potentiam Ltd

Timeline: May 2025 - June 2025

Work done:

  • Migrated the company’s HubSpot CRM reporting workflow from legacy SQL Server to PostgreSQL, significantly reducing infrastructure costs and improving refresh accuracy for sales analytics.
  • Engineered a robust Python-based ETL pipeline to extract data from HubSpot’s REST API, performing all cleaning, transformations, and loading into a Windows-hosted PostgreSQL instance.
  • Designed normalized PostgreSQL schemas for analytics-ready data, with efficient indexing and query optimization.
  • Automated incremental data loads and refreshes via cron jobs on Ubuntu, enabling Power BI dashboards to deliver near real-time insights.
  • Streamlined credential management and error logging for reliable unattended execution.
  • This enabled more frequent and accurate sales reporting, increasing value for the commercial team and reducing dependence on third-party services.

Skills & Technology:

  • Languages – Python, SQL.
  • Datastore – PostgreSQL (schema design, indexing, query optimization).
  • APIs – HubSpot API.
  • DevOps & Scheduling – Ubuntu server, Cron jobs, secure credential management.
  • BI & Reporting – Power BI (data modeling).
  • Workflow Orchestration – Scripted ETL orchestration, error handling, logging, alerting.

Private Enterprise GPT

Company: Potentiam Ltd

Timeline: Jan 2025 - Feb 2025

Work done:

  • Designed, configured, and deployed an in-house Private GPT system using an open source GitHub repository.
  • Hosted on an Ubuntu Linux server, secured with Nginx reverse proxy and SSL certification, and mapped to a custom domain.
  • Migrated user authentication system to OAuth with Microsoft Authenticator, restricting access to employees under the company domain.
  • Ensured resilience and automation by converting the application's backend into a system service, incorporating log rotation, cron jobs, and auto-restart mechanisms.
  • Converted the hardware setup containing GPU into a remotely accessible server via VNC, allowing seamless monitoring and usage by the entire company.
  • This deployment enabled the organization to replace external AI tools like ChatGPT and DeepSeek, ensuring data privacy by keeping all interactions in-house.

Skills & Technology:

  • Open Source Integration – Customizing an open source AI model to fit organizational needs.
  • OAuth Integration - Implemented Microsoft Authenticator-based login to allow only company domain users to access the platform.
  • DevOps & MLOps – Reverse proxy setup, SSL certification, domain mapping, service automation.
  • Database Management – SQLite configuration and data handling.
  • Hosting & Deployment – CI/CD principles, running application as a service, auto-restart & monitoring.
  • Linux Server Administration (Ubuntu) – System configuration for remote administration and usage.

AI Search

Company: Potentiam Ltd

Timeline: Dec 2024 - Jan 2025

Work done:

  • Developed an AI Search widget for the organisation's internal tool to provide smart and quick access to the employee salary data, market salary data, job specs, and skills matrices.
  • Enhanced the traditional search bar by integrating an “Ask AI” feature that processes user queries and provides accurate results with source links, cutting user navigation time by 50%.
  • Designed and implemented a RAG (Retrieval-Augmented Generation) system, incorporating:
    • FAISS for embedding storage and similarity search.
    • LangChain to orchestrate document retrieval, query formatting, and response generation using the LLM.
    • Optimized embedding algorithms and document-based metadata storage, improving accuracy from 75% to 90% and reducing hallucinations.
  • Introduced caching with SQLite to handle repeated queries efficiently, minimizing redundant computation.
  • Implemented structured HTML response validation to ensure consistent and interactive UI experience.
  • Deployed the production system via IIS, by bundling it into a JavaScript widget for seamless integration into the company’s website.

Skills & Technology:

  • Frontend Development - React.js, TailwindCSS
  • Backend Development - Flask (Python), MongoDB (user conversations), SQLite (cache)
  • RAG Architecture -
    • Embedding and similarity search: FAISS, PCA for dimensionality reduction
    • LLM: Ollama server with Llama3.x models
    • LangChain for end-to-end pipeline integration
    • Prompt templates and validation with Pydantic and BeautifulSoup
  • Deployment - IIS server hosting with optimized domain and site configuration
  • Problem Solving -
    • Enhanced embedding accuracy and response speed.
    • Reduced hallucinations with document-based metadata storage.
    • User-centric improvements validated by a 90% satisfaction rate and 50% time savings.

Intelligence Framework

Company: Potentiam Ltd

Timeline: Sept 2024 - Dec 2024

Work done:

  • Designed and developed an "All-in-One Intelligence Framework" to assist Talent Acquisition and HR teams with candidate evaluation, job specification creation, and data management.
  • Features Developed:
    • Resume Reviewer: Analyzed candidate resumes against job descriptions, providing strengths, weaknesses, and training suggestions. Results were downloadable and processed for up to 10 candidates simultaneously within 10 seconds.
    • General Chatbot: Built an LLM-powered chatbot with a dynamic UI for safe, secure company use. Features included conversational history, responsive design, and well-formatted chat messages.
    • Job Spec and Skills Matrix Generator: Created job specifications and skills matrices from job titles, trained strictly on internal company data, with outputs downloadable as Word documents.
    • Resume Formatter: Transformed pasted resume content into client-specific templates, ensuring formatting consistency without altering content. Each resume was stored in conversational history for easy reference.
    • Job Title Comparison: Enabled role comparison with drop-down menus, generating detailed analyses based on internal databases without internet-based data.
  • Boosted the TA team's efficiency by reducing candidate screening time from hours to seconds and reducing the HR team's workload by enabling instant internal data generation.

Skills & Technology:

  • Frontend Development - React.js, JavaScript, HTML, CSS
  • Backend Development - Flask (Python), MongoDB (storing user conversations)
  • AI Architecture -
    • LLMs: Ollama server for local LLM hosting and processing
    • LangChain: For context handling, prompt templates, and conversational workflows
  • Problem Solving -
    • Delivered fast (2-3 seconds) and accurate conversational AI responses.
    • Streamlined manual HR processes, significantly improving efficiency and reducing human errors.
    • Designed user-friendly and responsive UIs for seamless interaction and data download.

Chatbot Development

Company: Potentiam Ltd

Timeline: Jan 2024 - Sept 2024

Work done:

  • Spearheaded the development of an intelligent chatbot for Potentiam Ltd as a widget, leveraging an in-house trained Small Language Model (SLM) with internal company data.
  • Redesigned and developed a fully customized UI, integrating key functionalities:
    • File Downloads: Generating and downloading formatted Word and PDF outputs.
    • Speech-to-Text (STT): Allowing voice-based interaction.
    • Conversational Features: Enhanced handling of interruptions and context switching for smoother user interactions.
  • Developed custom actions (python functions) for data preparation, database setup, and Rasa integration for efficient data retrieval.
  • Enhanced chatbot intelligence by integrating LLMs (Ollama) for fallback responses when queries exceeded SLM's knowledge base, ensuring continuous and relevant interactions.
  • Demonstrated expertise in Natural Language Processing (NLP), Natural Language Understanding (NLU) through Entity and Intent recognition, and LLM tooling through the project's architecture.
  • Migrated the chatbot from Rasa Open Source to Rasa Pro CALM, enabling advanced conversational AI with fallback handling and seamless context-switching.

Skills & Technology:

  • Full Stack Development -
    • Frontend: HTML, CSS, JavaScript (Open source widget, later migrated to in-house UI developed using React.js)
    • Backend: Flask (API calls), MongoDB (knowledge base)
  • AI Architecture -
    • Rasa: Migrated from Open Source to Rasa Pro CALM
    • LLM Tooling: Integrated LLMs (Ollama) for fallback scenarios
  • Problem Solving -
    • Debugged installation issues, implemented custom solutions for data retrieval, and improved conversational design.
    • Designed scalable systems for handling diverse and dynamic data inputs without compromising response times.

Organisation Chart Generator

Company: Potentiam Ltd

Timeline: Jun 2024 - Jul 2024

Work done: Developed two key applications designed to automate and visualize organizational hierarchy efficiently.

  • Excel Generator:
    • Utilized an in-house Automated Web Scraping tool to extract relevant data and organize it into a downloadable Excel sheet.
    • The tool was containerized using Docker, ensuring seamless deployment and scalability.
    • UI Development was achieved with Flask and Python, offering a user-friendly interface for data extraction and download.
    • Technologies used: Beautiful Soup, ChromeDriver, Docker.
  • Chart Generator:
    • Took the preprocessed Excel file from the Excel Generator as input.
    • Produced high-quality hierarchy graphs in PNG format for download.
    • Built using the Graphviz Python library for graph visualization.
    • Hosted as a secure site on the company's IIS server, ensuring accessibility within the corporate domain.
    • Technologies used: Flask, Python, Graphviz, IIS Hosting.

This project enabled Sales and Marketing teams to generate organizational charts in under 15 seconds, vastly improving efficiency. The tool quickly became the most utilized resource for these teams and remains under continuous development to handle the latest package updates and maintain compliance with web security standards.

Skills & Technology:

  • Frontend and Backend Development - Flask, Python, HTML, CSS, MongoDB (User login detials)
  • Data Processing and Visualization - Beautiful Soup, ChromeDriver, Graphviz
  • DevOps - Docker for containerization, IIS Hosting
  • Problem Solving - Boosted operational efficiency, automation of data processing, and integration of secure deployment practices

Q&A system over documents

Company: Wipro Ltd

Timeline: May 2023 - June 2023

Work done: Fine-tuned LLMs using LangChain and implemented a simple Q&A system over documents, optimizing its accuracy, and created an interface using Streamlit. This work enabled users to have interactive and informative conversations while effortlessly extracting valuable insights from large volumes of textual data.

Skills & Technology: LangChain | Streamlit | ChatGPT API

Project Management

Web Applications Development

Company: Potentiam Ltd

Timeline: Dec 2023 - Jun 2025

Work done: Led the coordination between management and development teams for two different web applications tailored for internal staff as well as the clients. Oversaw client requirement gathering, workflow design, data management, testing and maintainance phases.

Skills & Technology: Project Management | Team Communication | Stakeholder requirments | Data processing | Full Stack: Angular, RESTful APIs, MySQL

Mentoring & Guidance

Internship Mentor II

Company: Potentiam Ltd

Timeline: Aug 2025 - Nov 2025

Work done: Guided a pre-final year Computer Science intern in building and deploying a RAG-based system. The system leveraged LangChain as its core framework, Qdrant for vector storage, Gemini API and Ollama for embedding, retrieval, and the LLM, and Streamlit for the UI, which was hosted on the community server.

Skills & Technology: Python | RAG | NLP | Project Management | Mentorship | Communication | Technical Assistance | Deployment | Git

Internship Mentor IExternal Link

Company: Potentiam Ltd

Timeline: Sept 2024 - Nov 2024

Work done: Mentored a pre-final year AI/ML Student intern on a Live Video-based Face Mask Detector project. The system leveraged OpenCV for video stream processing along with TensorFlow and Keras libraries for mask detection, based on Haar Cascade Classifier which was trained in-house using the raw data provided.

Skills & Technology: Python | Computer Vision | Project Management | Mentorship | Communication | Technical Assistance

Other projects

Cybersecurity Audit

Company: Potentiam Ltd

Timeline: Nov 2023 - Jan 2024

Work done: As part of a comprehensive cybersecurity audit within my organization, I actively participated in evaluating IT operations and assessing the effectiveness of cybersecurity measures. My responsibilities included conducting thorough checks on firewall configurations, antivirus software, and internal operating system settings. Utilizing tools such as Qualys, I identified vulnerabilities and implemented resolutions to enhance the security posture of the organization.

Additionally, I performed in-depth analyses of network settings and utilized Windows Local User and Group Management (lusrmgr) alongside the Registry Editor to ensure compliance with security policies and best practices.

Skills & Technology: Cybersecurity Assessment | Vulnerability Identification | Antivirus Management | Operating System Security | Network Configuration Analysis | User and Registry Management | Techs: Qualys, Windows Firewall, Windows OS Tools including lusrmgr (Local Users and Groups) and Registry Editor

Power BI Dashboards

Company: Potentiam Ltd

Timeline: Oct 2023 - Dec 2023

Work done: Developed Dashboards using PowerBI which provided an in-depth and smart overview of data for the Finance, Sales and Marketing teams respectively.

Skills & Technology: PowerBI | Data Preprocessing | Data Analysis