Muhammad Mohsin Badar
ML Engineer
8+ years exp
6+ years ML exp
Python
Node JS
SQL
JavaScript
Active 2 days ago
Download CVCV
Overview
Technical skills
Projects
Timeline
Roles
Overview
LLM engineer focused on agentic RAG workflows and SQL-generating LLM tools at a senior level with pragmatic system design. The strongest proven skill is building agent orchestration and RAG pipelines, evidenced by langchain_langgraph_RND/langgraph/gemini/langchain_gemini_rag_sql.py and ai_resume_screener/agents/nodes.py. There is little to no evidence of custom model training, systematic experiment tracking, or production deployment automation in the public code.
Technical skills
Languages
4
Python
Node JS
SQL
JavaScript
Python
7
Flask
FastAPI
Django
Requests
Pydantic
Boto3
HTTPX
Databases
6
MySQL
Redis
PostgreSQL
Apache Kafka
FAISS
Chroma
DevOps
13
GitHub
GitHub Actions
Amazon EC2
Rest API
Docker
Grafana
CI/CD
Amazon S3
AWS Lambda
Prometheus
Kubernetes
GCP
AWS
AI/ML
13
LLM
OpenCV
LangChain
Guardrails AI
Machine Learning
LangGraph
Pydantic AI
Prompt Engineering
RAG
OpenAI
LangSmith
huggingface_hub
Model Context Protocol
Other
25
Firebase
Express
Oracle
NumPy
Pandas
Scikit-learn
YOLO
PyTorch
TensorFlow
WebSockets
Context Engineering
Matplotlib
Databricks
Spark
LLMOps
MQTT
Databases
AI Agents
LLM Apps
AI/ML
Google ADK
LLM Guardrails
OpenRouter
Computer Vision
Microsoft Excel
Projects
Designed an ANPR system for vehicle license plate detection and text recognition, integrated with CCTV and databases. Implemented image processing and a detection/tracking pipeline to support plate extraction in real-world environments. Exposed results through RESTful services for downstream parking management workflows.
Financial AI Assistant
Built a personalized financial assistant using RAG with agentic workflows and context engineering. Integrated a state-backed API layer for real-time financial data ingestion and designed retrieval/response evaluation for reliability. Added monitoring and QA automation to support production-ready AI responses.
AI Based Safety Enhancement
Implemented a real-time computer-vision pipeline for detecting humans on CCTV streams using a YOLO-based detector. Designed a modular architecture for streaming, detection, and display components, including low-latency communication between processes. Improved real-time scalability by processing multiple camera streams concurrently.
Outlet AI Footfall Analysis
Created a person detection and tracking pipeline for CCTV footage to generate track IDs for individual detections. Added business logic to count people by categories (entrance, exit, and pass-by) based on tracker outputs. Exposed the aggregated results through RESTful APIs and persisted data in a database.
AI-IoT Based Security System
Developed an AI-driven premises and fleet security system combining computer vision with IoT sensor data. Built scalable backend services that connect frontend applications with smart devices and support real-time threat detection. Implemented automated alerting and emergency workflows using APIs and push/SMS/email notifications.
Timeline
AI/ML Engineer
·
Middle
Evren AI
·
Full-Time
Designed backend services and microservices to support multi-agent and agentic AI workflows. Built production AI features including RAG architecture, LLM workflow traceability, and LLM evaluation/QA automation. Implemented semantic search and alert/notification services, and maintained data pipelines with CI/CD and scalable DevOps practices on AWS and GCP.
Chroma
PostgreSQL
Python
FastAPI
GitHub
Docker
GCP
Amazon EC2
AWS Lambda
Amazon S3
Pydantic AI
Guardrails AI
LangChain
Redis
LangGraph
LangSmith
CI/CD
GitHub Actions
Rest API
WebSockets
Prompt Engineering
Context Engineering
RAG
Model Context Protocol
LLMOps
LLM
Machine Learning
Senior Software Engineer
·
Senior
Digital Myth Solutions
·
Full-Time
Built large-scale backend systems using microservices, data pipelines, and RESTful service integrations. Owned CI/CD and GitHub Actions workflows and supported cloud deployments across AWS and Oracle Cloud. Implemented AI capabilities using LLM-related tooling (prompt/context engineering, Guardrails, and agent frameworks) and added observability via Kafka, Grafana, Prometheus, and Rollbar.
PostgreSQL
Python
FastAPI
Django
GitHub
Docker
Kubernetes
Amazon EC2
AWS Lambda
Amazon S3
Pydantic AI
Guardrails AI
LangChain
Firebase
Redis
LangGraph
Apache Kafka
Grafana
Prometheus
CI/CD
GitHub Actions
Rest API
Prompt Engineering
Context Engineering
LLM
Machine Learning
Software Engineer
·
Middle
Bricks and Mortar-AI.tech
·
Full-Time
Developed backend services and RESTful APIs with database design and data pipeline implementation. Delivered cloud deployments with AWS and managed CI/CD automation using GitHub Actions. Contributed to AI/ML and computer-vision workflows, including data processing, model fine-tuning, and production monitoring with Grafana and Prometheus.
MySQL
Python
Flask
OpenCV
Node JS
Express
GitHub
Docker
Amazon EC2
Firebase
Redis
Grafana
Prometheus
CI/CD
GitHub Actions
Rest API
Machine Learning
Software Engineer
·
Middle
Aletheia-AI.tech
·
Full-Time
Worked on full-stack backend responsibilities such as API routing, server-side scripting, and database management. Implemented CI/CD automation using GitHub Actions and maintained RESTful API services. Contributed to software testing and development of reliable service integrations for AI-related applications.
MySQL
Python
Flask
GitHub
CI/CD
GitHub Actions
Rest API
Senior AI/ML Engineer
Confidence: High LLM Engineer
LLM engineer focused on agentic RAG workflows and SQL-generating LLM tools at a senior level with pragmatic system design. The strongest proven skill is building agent orchestration and RAG pipelines, evidenced by langchain_langgraph_RND/langgraph/gemini/langchain_gemini_rag_sql.py and ai_resume_screener/agents/nodes.py. There is little to no evidence of custom model training, systematic experiment tracking, or production deployment automation in the public code.
Model Architecture & Training
2/10
How well models are designed and trained
No custom model architectures or training loops; code configures and selects hosted/text LLMs, sets temperature, and uses structured-output wrappers around provider clients.
Evidence
ai_resume_screener/agents/models.py: get_text_model returns ChatOllama or ChatOpenAI configured from config.py
ai_resume_screener/config.py: LLM_MODEL_NAME and LLM_TEMPERATURE configuration and provider selection
Data Pipeline & Feature Engineering
3/10
How data is prepared for models
Basic data handling and schema-to-document splitting for RAG, simple resume file I/O and CSV report generation; no large-scale ETL, feature pipelines, or augmentation pipelines.
Evidence
langchain_langgraph_RND/langgraph/gemini/langgraph_gemini_rag_sql.py: split_schema_by_table and create_vector_db building FAISS from Documents
ai_resume_screener/agents/nodes.py: load_job_description, extract_resume_text, and compile_results manage resume files and CSV reports
Experimentation & Evaluation
2/10
How results are measured and tested
Lightweight validation and test harnesses (in-memory DB validation for generated SQL) and sensible CLI printing; no experiment tracking, reproducible runs, or systematic A/B/evaluation frameworks.
Evidence
langchain_langgraph_RND/langgraph/gemini/langgraph_gemini_lightrag_sql_test.py: create_in_memory_db and test_sql_query_node validating SQL via sqlite
ai_resume_screener/main.py: batch-run loop that prints results and handles per-resume exceptions
MLOps & Deployment
3/10
How models are shipped to production
Graph-based orchestration (LangGraph), environment-driven config, and file-based outputs; lacks production-grade deployment infra, CI/CD, monitoring, or model/version management tooling.
Evidence
ai_resume_screener/workflows/graph.py: StateGraph setup and compiled app
ai_resume_screener/config.py: environment-driven paths and config constants
Computational Efficiency
2/10
How efficiently computing resources are used
Uses FAISS and HuggingFace embeddings for vector search and lists GPU-enabled dependencies in another project, but no explicit efficiency engineering (quantization, batching, memory profiling, or distributed training) is implemented.
Evidence
langchain_langgraph_RND/langgraph/gemini/langgraph_gemini_rag_sql.py: create_vector_db uses HuggingFaceEmbeddings and FAISS
langchain_langgraph_RND/requirements.txt: faiss-cpu and torch listed (dependency evidence)
Research Depth & Innovation
3/10
Depth of research and new ideas
Shows pragmatic engineering patterns for agentic LLM workflows and LightRAG-style prompting with retries and in-memory validation, but no novel algorithms or formal research contributions.
Evidence
langchain_langgraph_RND/langgraph/gemini/langgraph_gemini_lightrag_sql.py: LightRAG prompt construction and graph node for SQL generation
langchain_langgraph_RND/langgraph/gemini/langchain_gemini_lightrag_sql_test.py: retry logic and conditional edges to reattempt SQL generation after failures
Expertise
AI / LLM Engineering (Agents)· Senior
RAG· Senior
Industries
Artificial Intelligence· Middle
Technologies
SQL
MySQL· since 2018
Redis· since 2020
LangGraph· since 2024
LangChain· since 2023
Spark
GCP· since 2025
Databricks
Oracle
Chroma· since 2025
OpenCV· since 2020
Model Context Protocol· since 2025
YOLO
GitHub Actions· since 2018
Prometheus· since 2020
WebSockets· since 2025
Scikit-learn
Prompt Engineering· since 2023
Computer Vision· since 2026
AI Agents
Guardrails AI· since 2023
LangSmith· since 2025
Pydantic AI· since 2024
CI/CD· since 2018
TensorFlow· since 2026
Pandas
NumPy
PyTorch
AWS
Docker· since 2020
Kubernetes· since 2023
Grafana· since 2020
Google ADK
AWS Lambda· since 2023
Amazon EC2· since 2020
OpenRouter
LLMOps· since 2025
GitHub· since 2018
Amazon S3· since 2023
Context Engineering· since 2025
LLM Guardrails
Machine Learning· since 2020
Recommendations
- Develop RAG-backed agent services that convert natural language to validated SQL and integrate them with vector stores and FAISS for schema-aware retrieval.
- Build robust evaluation and experiment tracking (MLflow or W&B) for prompt/temperature ablations and structured-output accuracy measurement.
- Implement secure secret handling and CI/CD for on-prem or cloud deployments (avoid hardcoded keys; add automated tests and containerization).
- Add efficiency work such as quantization or batching strategies and end-to-end latency monitoring if aiming for production-grade LLM services.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Data Scientist
Confidence: Medium Data Engineer
LLM-focused data engineer (mid-level) who builds agentic LLM pipelines and RAG-driven SQL/assistant tooling. The strongest proven skill is designing small state-graph orchestration and LLM-to-SQL workflows as shown by the LangGraph-based graph nodes and SQL generator/testing harness (langgraph_gemini_lightrag_sql.py and langgraph_gemini_lightrag_sql_test.py). There is little evidence of statistical modeling, production-scale distributed data pipelines, formal test coverage or observability and security hardening in production.
Statistical Rigor
Correct use of statistics
Not evidenced in public code
Data Wrangling & Cleaning
5/10
Preparing and cleaning data
Practical file-based ETL and defensive data handling is implemented: safe path resolution, resume reading, structured LLM output coercion, CSV reporting and resume archival. The code shows attention to input validation and defensive defaults but is limited to small-scale, local file workflows rather than large-scale data engineering concerns.
Evidence
ai_resume_screener/agents/nodes.py: extract_resume_text and compile_results implement resume parsing, CSV reporting and safe file moves
ai_resume_screener/config.py: _path and centralized path constants (RESUME_DIR, REPORT_FILE) with .env loading
Exploratory Analysis & Visualization
Exploring and visualizing data
Not evidenced in public code
Predictive Modeling
Building models that predict
Not evidenced in public code
Business Insight & Impact
4/10
Turning analysis into business value
Basic business reasoning is present: candidate scoring, a pass threshold and durable decision logic executed in Python rather than trusting the model. The work is focused and pragmatic for HR screening but lacks deeper cost-of-error analysis or stakeholder metrics.
Evidence
ai_resume_screener/agents/nodes.py: score_candidate clamps and thresholds model output and decides pass/fail
ai_resume_screener/config.py: PASS_SCORE_THRESHOLD centralizes the business rule
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Reproducibility basics exist such as pinned requirements, .env-driven config, and a single config module to avoid hardcoded paths. There is limited evidence of CI, data versioning, seeds, or extensive test coverage.
Evidence
ai_resume_screener/requirements.txt: pinned dependencies list
ai_resume_screener/config.py: dotenv loading and centralized settings used across modules
ai_resume_screener/main.py: single-run orchestration reuses parsed job description for deterministic runs
Expertise
Data Science· Middle
Industries
Artificial Intelligence· Middle
Software· Middle
Technologies
Databases
AI/ML
LLM Apps
Python· since 2018 · Senior
PostgreSQL· since 2023
FAISS
huggingface_hub
LLM· since 2023
RAG· since 2025
OpenAI
Recommendations
- Develop RAG-powered SQL assistants and LLM orchestration pipelines that validate generated SQL with in-memory or sandboxed DB tests, building on the existing langgraph_* sql generator and test harness.
- Build offline LLM screening pipelines that add support for non-text resume inputs (PDF and DOCX parsing), schema validation, and richer provenance metadata before moving resumes.
- Productionize the agent graphs by adding CI tests, instrumentation and checkpointing, secrets management for API keys, and a deployment plan for scale and multi-user isolation.
- Integrate vector DB storage and retrieval evaluation (metric tracking for retrieval quality) and add routine error analysis to quantify failure modes of LLM-generated SQL.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Backend Developer
Confidence: Medium Generalist
LLM-focused Python backend engineer (Middle level) who builds small, agentic screening pipelines and orchestration for LLM-driven workflows. The strongest proven skill is architecting an LLM-based screening pipeline as implemented in ai_resume_screener (agents/nodes.py, workflows/graph.py, agents/models.py). There is limited evidence of production APIs, databases/migrations, testing, concurrency controls, or observability/metrics in the public code.
API Design
2/10
How well APIs are designed
Minimal API design evidence; this codebase contains an offline batch runner and a state graph but no HTTP API contracts, versioning, or idempotency design.
Evidence
ai_resume_screener/main.py: batch runner invoking the graph rather than exposing an HTTP API
ai_resume_screener/workflows/graph.py: StateGraph wiring (no REST/HTTP layer or API contract)
Data Layer & Database
2/10
Working with databases
Very lightweight data layer using CSV and filesystem operations; no database, migrations, transaction handling, or tuned queries are present.
Evidence
ai_resume_screener/agents/nodes.py: compile_results writes CSV rows to reports/screening_results.csv
ai_resume_screener/agents/nodes.py: moves resume files between shortlisted/ and rejected/ folders using shutil.move
ai_resume_screener/config.py: central path helpers and environment-driven file locations
Scalability & Performance
3/10
Handling load and speed
Some basic scalability thinking is visible (parse job description once, single model instances), but no queueing, caching strategies with invalidation, connection pooling or measured performance work.
Evidence
ai_resume_screener/agents/nodes.py: load_job_description is documented and intended to run once per run to avoid repeated parsing
ai_resume_screener/main.py: reuses the parsed job description across all resume runs
ai_resume_screener/agents/models.py: single instantiation of model clients and provider selection in get_text_model
System Architecture
4/10
Overall system structure
Clear modular structure and separation of concerns (config, agents, models, workflows/state graph); the state-graph orchestration is a deliberate architectural choice for pipeline composition.
Evidence
ai_resume_screener/config.py: centralized configuration and path resolution
ai_resume_screener/workflows/graph.py: clear graph nodes and edges (extract -> score -> compile)
ai_resume_screener/agents/nodes.py: encapsulated node implementations for each pipeline step
Security & Auth
2/10
Protecting data and access
Basic boundary checks and environment-driven secret loading exist, but there is no advanced auth, token lifecycle, input sanitization beyond file-existence checks, or explicit secrets management practices.
Evidence
ai_resume_screener/config.py: reads OPENAI_API_KEY and other secrets from environment via dotenv
ai_resume_screener/agents/nodes.py: FileNotFoundError checks and clamping of model-returned candidate_score
Reliability & Observability
3/10
Stability and monitoring
Some defensive and reliability practices are present (defensive defaults, clamping, try/except per-resume, safe file overwrite behavior), but no structured logging, retries/backoff, timeouts or observability instrumentation.
Evidence
ai_resume_screener/agents/nodes.py: defensive defaults for structured model outputs and clamped candidate_score
ai_resume_screener/main.py: per-resume try/except to prevent one failure from stopping the batch
ai_resume_screener/agents/nodes.py: checks and overwrite behavior when moving files (removes existing dest before move)
Expertise
Backend AI & LLM· Middle
Python· Middle
Industries
Professional Services· Middle
Technologies
Node JS· since 2020 · Junior
Rest API· since 2018
Flask· since 2018
Express· since 2020
FastAPI· since 2023
Django· since 2023
Apache Kafka· since 2023
Pydantic
HTTPX
Requests
Boto3
Recommendations
- Build and productionize small LLM orchestration services that expose well-defined HTTP APIs and idempotent endpoints (start by adding a FastAPI wrapper with request validation and idempotency keys).
- Implement persistent storage and migrations for results (replace CSV/files with a transactional database or vector DB if adding embeddings) and add simple migration history.
- Add observability and reliability features: structured logging with correlation ids, retries with backoff for external calls, model timeouts, and basic metrics/alerts.
- Harden security and secrets: avoid plaintext env files in production, add token lifecycle handling if exposing APIs, and validate/sanitize model inputs and outputs.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: