Suraj Paliwal - QA Engineer

Overview

A Middle-level Python ML/GenAI engineer focused on lightweight RAG chat applications and embedding-based search. The strongest proven skill is building end-to-end RAG prototypes combining BM25, dense embeddings and a vector store as implemented in search_engine.py and the build_context logic. Public code shows limited production hardening - secrets are in source, there are no tests, and operational concerns like retries, structured logging and CI are not evidenced.

Phone

Timeline

QA Automation Engineer · Middle

Secure Meters Ltd. · Full-Time

Jul 2022 to Present 4 Years 4 Months Udaipur In office

Designed and automated test plans for web and mobile applications using Python and PyTest to reduce manual testing effort across releases. Built and maintained automated test scripts for system integrations spanning API, data/ETL processes, and database layers. Developed Jenkins-based CI/CD pipelines for automated test execution, and performed API testing with Postman and JMeter to verify end-to-end flows and catch defects pre-release.

Python

Pytest

Postman

JMeter

Rest API

CI/CD

Jenkins

QA Automation Engineer (Trainee / Junior) · Junior

Secure Meters Ltd. · Full-Time

Jan 2022 to Jun 2022 5 Months Udaipur In office

Automated parts of data synchronization workflows using Python to improve operational efficiency and reduce manual errors. Supported build deployments by validating functionality with both manual and automated approaches across web, mobile, and API surfaces.

Python

Projects

ATS Resume Score Calculator

Developed an ATS resume scoring tool to evaluate resumes against required skills and job descriptions. Helped candidates understand how well their resume matches target keywords used by applicant tracking systems. Implemented logic to compute and present score results for comparison.

Scalable Test Framework with Excel Reporting

Designed a scalable automated testing framework with Excel-based reporting integration. Implemented automated JIRA-Excel workflows to support QA dashboards and trend analysis. Used the framework to standardize reporting for regression and functional coverage.

CI/CD Automation with Jenkins and Python

Created and optimized Jenkins CI/CD pipelines using Python scripts to automate test execution across build stages. Improved test run efficiency and reduced deployment validation time. Maintained stable automation behavior throughout the pipeline flow.

AMISP End-to-End Automation

Implemented end-to-end automation for AMISP workflows to validate data integrity across multiple layers. Covered functional and regression checks across API, web application, and database components. Focused on ensuring end-to-end verification from upstream inputs to stored outputs.

LLM-Based Smart Chatbot

Built an LLM-powered chatbot to answer system and ticket-related questions using natural language processing. Integrated JIRA data and implemented secure document-based Q&A. Used LangChain and FastAPI to connect the chatbot logic to the application interface.

Technical skills

Languages

3

Python

SQL

Bash

QA

4

Pytest

Postman

JMeter

Playwright

DevOps

4

Rest API

CI/CD

Jenkins

Azure

AI/ML

4

Pandas

NumPy

Streamlit

LLM

Other

25

Requests

PostgreSQL

FastAPI

Selenium

LangChain

NLTK

Redis

Appium

Robot Framework

Azure SQL Database

Qdrant

MS SQL

Power BI

GitHub

Netlify

Docker

Git

Linux

Windows

Azure Data Factory

HAProxy

QA

Zabbix

NLP

ETL/ELT

Middle AI/ML Engineer Confidence: Medium LLM Engineer

LLM-focused engineer (mid-level) specializing in retrieval-augmented chatbots and document ingestion pipelines. The strongest proven skill is building an end-to-end embedding and retrieval pipeline for document/Jira data as implemented in data_loader.py and search_engine.py. There is no evidence of custom model training, systematic experiment tracking, production-grade deployment, or meaningful performance optimization in public code.

Model Architecture & Training

1/10

How well models are designed and trained

No custom model architectures or training loops; uses a pretrained SentenceTransformer for embedding only.

Evidence

main.py: load_model() using SentenceTransformer("all-MiniLM-L6-v2")

data_loader.py: process_uploaded_file() calls model.encode(...) to create embeddings

Data Pipeline & Feature Engineering

4/10

How data is prepared for models

Reasonable end-to-end data ingestion and cleaning pipeline with chunking, field mapping and DB insert logic for document/Jira data.

Evidence

data_loader.py: FIELD_MAP and DB_COLUMNS mapping plus column normalization and chunking logic

data_loader.py: clean_value(), safe_text() helpers used during record preparation

data_loader.py: execute_values(...) insert into jira_tickets (DB insert pipeline)

Experimentation & Evaluation

1/10

How results are measured and tested

Minimal experimentation and evaluation artifacts; only basic debug output in the UI, no experiment tracking or validation suites.

Evidence

main.py: st.expander('🔍 Debug Logs') for ad-hoc logs

llm_client.py: inline comment describing HF API parameters (no tracking or evaluation harness)

MLOps & Deployment

2/10

How models are shipped to production

Simple serving approach (Streamlit) and vector DB integration (Qdrant), but no deployment, versioning, monitoring or CI/CD evidence.

Evidence

main.py: Streamlit app orchestrating search + LLM calls

search_engine.py: hybrid_search() using get_client() (Qdrant) to retrieve vectors

qdrant_store.py: get_client() import of QdrantClient (client integration for vector DB)

Computational Efficiency

1/10

How efficiently computing resources are used

No explicit efficiency work; embeddings and searches are performed naively per-row without batching, quantization or GPU profiling.

Evidence

data_loader.py: per-row model.encode(...) in process_uploaded_file()

search_engine.py: simple client.search(...) without batching or advanced query options

Research Depth & Innovation

1/10

Depth of research and new ideas

No research or paper-level implementations; the code implements standard retrieval + prompt flow without novel algorithms.

Evidence

llm_client.py: HF inference prompt wrapper implementing a simple generation call

search_engine.py: straightforward conversion of qdrant results into context blocks

Expertise

RAG· Middle

Conversational AI & Chatbots· Middle

Industries

Information Technology· Middle

Software· Middle

Technologies

Python· since 2022 · Middle

SQL

PostgreSQL

Redis

LangChain

Zabbix

Qdrant

NLP

Azure

MS SQL

HAProxy

CI/CD· since 2022

Jenkins· since 2022

Pandas

NumPy

Git

Docker

LLM

NLTK

Streamlit

Requests

Azure SQL Database

GitHub

Linux

Windows

Machine Learning· mentioned only

Recommendations

Develop internal knowledge-base chatbots or RAG prototypes that ingest documents and provide contextual answers using the existing embedding + Qdrant pipeline.
Build Streamlit prototype frontends and iterate on data ingestion and chunking logic for document QA; focus on input validation and secure secrets handling.
Harden the pipeline for production: add batching, retry/backoff for external calls, unit tests, CI/CD and basic monitoring for model and vector DB health.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Build-1.0.0.2_chatbot Data-science-journey

Middle AI / LLM Engineering (Agents) Confidence: Medium Generalist

A Middle-level Python ML/GenAI engineer focused on lightweight RAG chat applications and embedding-based search. The strongest proven skill is building end-to-end RAG prototypes combining BM25, dense embeddings and a vector store as implemented in search_engine.py and the build_context logic. Public code shows limited production hardening - secrets are in source, there are no tests, and operational concerns like retries, structured logging and CI are not evidenced.

Architecture & Organization

4/10

How the project is structured

Modular app-level organization with separate UI, ingestion, search and LLM client modules, but the architecture is application-prototype quality rather than production-grade system design.

Evidence

Build-1.0.0.2_chatbot/main.py: orchestrates Streamlit UI, RAG search and single LLM call

JIRA_CHAT_BOT/search_engine.py: separates BM25, dense search and RRF fusion into reusable functions

Build-1.0.0.2_chatbot/data_loader.py: ingestion + chunking + upsert to vector store (qdrant) as a distinct component

Code Quality & Standards

3/10

How clean the code is

Readable Python code and modular files but inconsistent error handling, duplicated code, hardcoded config and no tests or typing.

Evidence

Build-1.0.0.2_chatbot/data_loader.py: duplicated try/except blocks and mixed in-memory vs DB logic

Build-1.0.0.2_chatbot/llm_client.py: simple HF integration with a single broad try/except and a generic fallback

JIRA_CHAT_BOT/Jira_ChatBot.py: UI, DB schema creation and ingestion logic mixed in one file

Engineering Culture

2/10

Teamwork and good practices

Some attention to UX (Streamlit themes and debug expander) but little in the way of engineering process artifacts like tests, CI, CONTRIBUTING or changelogs.

Evidence

Build-1.0.0.2_chatbot/main.py: debug logs expander and use of st.cache_resource showing some operational thinking

JIRA_CHAT_BOT/README.md: minimal README present (documentation is sparse)

Technical Complexity

4/10

How hard the problems solved are

Contains non-trivial information-retrieval logic (BM25 + dense + RRF fusion and context prioritization) showing application-level algorithmic work but not advanced scalability or novel algorithms.

Evidence

JIRA_CHAT_BOT/search_engine.py: bm25_search, dense_search, rrf_fusion and build_context implementations

Build-1.0.0.2_chatbot/search_engine.py: compact qdrant-based dense search and context builder for RAG

Security & Risks

2/10

Protection and risk handling

Direct security risks are present (hardcoded credentials, broad exception handling); some secrets are read via st.secrets but sensitive data appears in source.

Evidence

JIRA_CHAT_BOT/Jira_ChatBot.py: production DB host and plaintext credentials in source

Build-1.0.0.2_chatbot/data_loader.py: DB_PASSWORD hardcoded as 'Secure@123'

Build-1.0.0.2_chatbot/llm_client.py: prompt contains explicit operational rules (prompt-injection content embedded in code)

Reliability & Fault Tolerance

2/10

Stability when things fail

Basic try/except usage prevents crashes but there is minimal structured logging, no retries/backoff, and no test harness or monitoring hooks.

Evidence

Build-1.0.0.2_chatbot/data_loader.py: broad try/except blocks that return generic Streamlit errors

Build-1.0.0.2_chatbot/llm_client.py: single try/except with a generic user-facing fallback and no retry logic

Industries

Artificial Intelligence· Middle

Software· Middle

Recommendations

Develop retrieval-augmented chatbot prototypes and internal knowledge assistants using Streamlit, SentenceTransformers and Qdrant.
Build ingestion pipelines that produce chunked embeddings and upsert into vector stores, focusing on data normalization and chunk-priority logic.
Prototype hybrid search features (BM25 + dense + fusion) and iterate on context building for LLM prompts.
Harden prototypes for production by removing hardcoded secrets, adding parameterized DB access, structured logging and automated tests.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Build-1.0.0.2_chatbot JIRA_CHAT_BOT

Junior QA Engineer Confidence: Low Generalist

A junior-level automation and scripting practitioner focused on small Python utilities and pragmatic automation tasks. The strongest proven skill is building a multithreaded Python bulk-email sender and file parsers that use ThreadPoolExecutor, Lock, pdfplumber and pandas to extract and process contact data. There is no evidence in the analyzed human-authored files of formal test automation frameworks, CI integration, contract testing or negative-path test design.

Test Automation Frameworks

Building automated tests

Not evidenced in public code

Test Coverage & Strategy

What and how to test

Not evidenced in public code

API & Integration Testing

Testing how parts work together

Not evidenced in public code

Performance & Load Testing

Testing speed under load

Not evidenced in public code

Bug Reporting & Analysis

Finding and describing bugs

Not evidenced in public code

CI Test Integration

Running tests automatically

Not evidenced in public code

Technologies

QA

Selenium

JMeter· since 2022

Playwright

Appium

Postman· since 2022

Pytest· since 2022

Robot Framework

Recommendations

Build internal automation utilities and small integration scripts that parse files and integrate with SMTP or other messaging systems, applying safety checks and secrets management.
Harden and extend the script into a testable module with unit tests, explicit error handling, configuration via environment variables, and secret management (no hardcoded credentials).
Develop small QA automation tasks that demonstrate test infrastructure skills such as Playwright test fixtures, negative-path API tests, and CI pipeline integration.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

capi-qa-assessment Auto_HR-Connect

Job preferences

Work permits: India