Commit ·
7adb4f3
0
Parent(s):
data creating and loading done
Browse files- .gitignore +2 -0
- Haider_Ali_AI_engineer_resume.md +91 -0
- __pycache__/dataloader.cpython-313.pyc +0 -0
- dataloader.py +16 -0
- main.py +27 -0
- pdf_2_markdown_converter.py +14 -0
.gitignore
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
data/
|
| 2 |
+
.env
|
Haider_Ali_AI_engineer_resume.md
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
## EDUCATION
|
| 2 |
+
|
| 3 |
+
## HAIDER ALI
|
| 4 |
+
|
| 5 |
+
Hali4@mail.yu.edu · https://www.linkedin.com/in/haiderzali/ · https://github.com/AliHaider20
|
| 6 |
+
|
| 7 |
+
Fairfax, VA · (551) 344-4769
|
| 8 |
+
|
| 9 |
+
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
|
| 10 |
+
|
| 11 |
+
Yeshiva University - NYC, USA -Master of Science, Artificial Intelligence, May 2024
|
| 12 |
+
|
| 13 |
+
Mumbai University - Mumbai, India - Bachelor of Science, Computer Science, Specialization in AI, May 2021
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
## PUBLICATIONS
|
| 17 |
+
|
| 18 |
+
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
|
| 19 |
+
|
| 20 |
+
- KatzBot : Revolutionizing Academic Chatbot for Enhanced Communication, May 2024
|
| 21 |
+
|
| 22 |
+
- Cow Teat Keratosis Classification : Data Augmentation Using Stable Diffusion for Imbalanced Biological Dataset, May 2023
|
| 23 |
+
|
| 24 |
+
## TECHNICAL SKILLS
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
|
| 28 |
+
|
| 29 |
+
- Languages & Frameworks: Python, C++, PyTorch, JAX, HuggingFace, Transformers, FastAPI, REST APIs, LangChain, LangGraph, Semantic Kernel, Streamlit, Flask, FastMCP
|
| 30 |
+
- MLOps, Cloud & Data: AWS, Azure, GCP (Vertex AI), Docker, Kubernetes, CI/CD, MLflow, LangSmith, ONNX, vLLM, Elasticsearch, Vector database, FAISS, PostgreSQL, MySQL, Pandas, HIPAA, EMR/EHR, FHIR, MCP
|
| 31 |
+
- ML & GenAI: LLMs (GPT, BERT, Gemini), RAG, Prompt/Context Engineering, NLP, Vector search (sparse, hybrid), Tool calling, Model evaluations, Fine-tuning (QLoRA, AWQ/GPTQ, SFT), FSDP, FlashAttention, Computer Vision, Stable Diffusion, XGBoost
|
| 32 |
+
|
| 33 |
+
## EXPERIENCE
|
| 34 |
+
|
| 35 |
+
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
|
| 36 |
+
|
| 37 |
+
Credo Health -Machine Learning Engineer; Austin, TX
|
| 38 |
+
|
| 39 |
+
June 2024 - Feb 2026
|
| 40 |
+
|
| 41 |
+
- Built a hybrid re-ranking RAG pipeline serving concurrent multi-user EMR/EHR queries with real-time document ingestion
|
| 42 |
+
- Reduced LLM prompt tokens by 30% via hash-based deduplication , lowering per-query inference cost at scale
|
| 43 |
+
- Deployed an agentic chatbot framework achieving 96% accuracy , validated across thousands of QA test cases
|
| 44 |
+
- Built a multi-judge LLM eval framework with physician golden sets running across all production agent outputs
|
| 45 |
+
|
| 46 |
+
S&P Global -Machine Learning Joint Research Project; NYC, NY
|
| 47 |
+
|
| 48 |
+
May 2023 - August 2023
|
| 49 |
+
|
| 50 |
+
- Scraped, engineered, and extracted relevant features of Global Precipitation Data from NASA's GDDP ( HDF ) files using R
|
| 51 |
+
- Employed predictive modeling using features from KNN and CNN after morphological image analysis
|
| 52 |
+
- Evaluated multiple super-resolution models; ESRGAN maintained 80% image quality while enhancing resolution
|
| 53 |
+
|
| 54 |
+
Kan Innovations -Machine Learning Engineer; Mumbai, India
|
| 55 |
+
|
| 56 |
+
## January 2022 - Feb 2023
|
| 57 |
+
|
| 58 |
+
- Developed foot imagery dataset using scrapers and pressure mat data, processed with Pix2Pix GANs
|
| 59 |
+
- Leveraged DenseNet121 , DeepLabV3+ , and custom segmentation model to estimate the pressure points in a foot
|
| 60 |
+
- Constructed 3D voxel foot meshes using Open3D library and Blender to help design insoles, reduced 40% foot deformity
|
| 61 |
+
- Implemented BlazePose model for real-time pose estimation, enhancing gaming interaction reducing 20% foot deformity
|
| 62 |
+
|
| 63 |
+
## PROJECTS
|
| 64 |
+
|
| 65 |
+
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
|
| 66 |
+
|
| 67 |
+
## [Dedup-RAG MCP Server | github.com/AliHaider20/MCP-VectorDB-context-manager | MCP | VectorDB](http://github.com/AliHaider20/MCP-VectorDB-context-manager)
|
| 68 |
+
|
| 69 |
+
July 2026
|
| 70 |
+
|
| 71 |
+
- Built a local-first RAG pipeline as an MCP server, integrating FAISS search and SQLite metadata storage for Claude
|
| 72 |
+
- Integrated jina-embeddings-v5 model with asymmetric query/document prompting for local, API-free retrieval
|
| 73 |
+
- Designed a two-layer dedupication system that cut 34% of chunks from re-embedding and reduced context by 10%
|
| 74 |
+
|
| 75 |
+
## Credit card assistant | github.com/AliHaider20/Semantic-kernel-card-assist-agent | Prompt engineering February 2025
|
| 76 |
+
|
| 77 |
+
- Developed agentic AI chatbot using Semantic Kernel with tool-calling for card operations and document Q&A
|
| 78 |
+
- Built RAG pipeline ingesting PDFs, chunking via LangChain, and indexing with FAISS for grounded responses
|
| 79 |
+
- Built reusable tool/skill abstractions with clear input/output interfaces for card operations and document retrieval
|
| 80 |
+
|
| 81 |
+
University Chatbot using LLM | github.com/AliHaider20/KatzBot-LLM | Paper | pre-training LLMs December 2023
|
| 82 |
+
|
| 83 |
+
- Scraped, and synthetically generated 60K question-answer pairs and human using the NLPAug tool, BERT , and GPT models
|
| 84 |
+
- Fine-tuned Phi 1.5B , Flan T5 , GPT 2 , and Mistral models using FSDP, QLoRA, SFT, and GRPO achieving 70% RougeL score
|
| 85 |
+
- Instruction trained, and deployed a custom SLM model on local server using Streamlit and FastAPI
|
| 86 |
+
|
| 87 |
+
## ACHIEVEMENTS
|
| 88 |
+
|
| 89 |
+
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
|
| 90 |
+
|
| 91 |
+
- 1st place, GreyAtom Data Science Competition; top performer in 2 consecutive Python Hackathons (700+ participants)
|
__pycache__/dataloader.cpython-313.pyc
ADDED
|
Binary file (1.29 kB). View file
|
|
|
dataloader.py
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from glob import glob
|
| 2 |
+
|
| 3 |
+
class ProfileLoader:
|
| 4 |
+
def __init__(self, data_dir="data"):
|
| 5 |
+
self.data_dir = data_dir
|
| 6 |
+
|
| 7 |
+
def load_and_format_markdown(self, filename):
|
| 8 |
+
data = ""
|
| 9 |
+
seperator = "-------"*20 + "\n"
|
| 10 |
+
md_files = glob(self.data_dir + "/*.md")
|
| 11 |
+
for md_file in md_files:
|
| 12 |
+
with open(md_file, "r", encoding="utf-8") as f:
|
| 13 |
+
data += f.read()
|
| 14 |
+
data += seperator
|
| 15 |
+
|
| 16 |
+
return data
|
main.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from langchain_core.prompts import ChatPromptTemplate
|
| 2 |
+
from langchain_groq import ChatGroq
|
| 3 |
+
from dataloader import ProfileLoader
|
| 4 |
+
from dotenv import load_dotenv
|
| 5 |
+
|
| 6 |
+
load_dotenv() # Load environment variables from .env file
|
| 7 |
+
|
| 8 |
+
profile_loader = ProfileLoader(data_dir="data")
|
| 9 |
+
# Load and format the Markdown content from the .md files
|
| 10 |
+
profile_data = profile_loader.load_and_format_markdown("data")
|
| 11 |
+
|
| 12 |
+
# 1. Define the system and user instructions
|
| 13 |
+
prompt = ChatPromptTemplate.from_messages([
|
| 14 |
+
("system", "You are an assistant that answers questions accurately based ONLY on the provided context."),
|
| 15 |
+
("user", "Context:\n<context>\n{context_text}\n</context>\n\nQuestion: {user_query}")
|
| 16 |
+
])
|
| 17 |
+
|
| 18 |
+
# 2. Format the prompt with your full Markdown string
|
| 19 |
+
llm = ChatGroq(model="openai/gpt-oss-120b", temperature=0, max_tokens=2048)
|
| 20 |
+
chain = prompt | llm
|
| 21 |
+
|
| 22 |
+
response = chain.invoke({
|
| 23 |
+
"context_text": profile_data,
|
| 24 |
+
"user_query": "What are the strongest and weakest skills of the candidate based on the provided context?"
|
| 25 |
+
})
|
| 26 |
+
|
| 27 |
+
print(response)
|
pdf_2_markdown_converter.py
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from docling.document_converter import DocumentConverter
|
| 2 |
+
import os
|
| 3 |
+
from tqdm import tqdm
|
| 4 |
+
|
| 5 |
+
file_paths = os.listdir("data")
|
| 6 |
+
|
| 7 |
+
for file_path in tqdm(file_paths):
|
| 8 |
+
if file_path.endswith(".pdf"):
|
| 9 |
+
doc_dir = "data"
|
| 10 |
+
converter = DocumentConverter()
|
| 11 |
+
doc = converter.convert(os.path.join(doc_dir, file_path)).document
|
| 12 |
+
|
| 13 |
+
with open(os.path.join(doc_dir, file_path.replace(".pdf", ".md")), "w", encoding="utf-8") as f:
|
| 14 |
+
f.write(doc.export_to_markdown())
|