Skip to content
View lucky-verma's full-sized avatar
💠
FTM
💠
FTM
23 followers · 52 following

Achievements

Achievement: Pull Sharkx2Achievement: Starstruck

Achievements

Achievement: Pull Sharkx2Achievement: Starstruck

Highlights

Organizations

@TandMcorp @Vast-Group-Backend @quniq

Block or report lucky-verma

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
lucky-verma/README.md

Typing SVG 

$ whoami
> Machine Learning Scientist building production agentic AI systems
> 5+ years: LLM orchestration, knowledge graphs, RAG, computer vision
> MS Computer Science @ UMBC (GPA: 3.97) | Fremont, CA

$ cat current_work.json
{
  "aegis": "Production agentic AI platform for defense supply chain security",
  "stack": "LangGraph + DSPy + Apache AGE + LanceDB + FastAPI",
  "scale": "Entity resolution across 10M+ corporate records",
  "domain": "Foreign ownership risk assessment (32 CFR 117.11)"
}

$ cat skills.json
{
  "ai_llm": ["RAG", "Agentic AI", "LangGraph", "DSPy", "vLLM", "Fine-tuning (LoRA/DPO)"],
  "ml_dl": ["PyTorch", "FSDP", "DeepSpeed", "SLURM Multi-GPU", "Triton Kernels", "CUDA"],
  "infra": ["Knowledge Graphs", "FastAPI", "Docker", "Kubernetes", "AWS", "PostgreSQL"]
}

$ ls -lh projects/
total 6 items

drwxr-xr-x  InfiniteContext-1B/          # 93.7% KV cache reduction, MLA architecture
drwxr-xr-x  Document-Classification/     # LayoutLM multimodal (35 stars, 13 forks)
drwxr-xr-x  US-Driver-License-Extraction/ # Production OCR with YOLO (25 stars)
drwxr-xr-x  information-retrieval/       # RAG: vector search + rerank + fine-tuned LLaMA 3
drwx------  AEGIS-Platform/              # 10M entities, agentic pipeline [proprietary]
drwx------  Research-Cockpit/            # 169 endpoints, 93K vectors [proprietary]

$ git log --oneline oss-contributions/
* HKUDS/LightRAG  — Apache AGE 1.5.0+ schema + SQL injection fix
* HKUDS/MiniRAG   — Docker crash fix

Tech Stack

languages: [Python, C++/CUDA, SQL, JavaScript]
frameworks: [PyTorch, LangChain, LangGraph, DSPy, HuggingFace]
infrastructure: [Docker, Kubernetes, FastAPI, AWS, PostgreSQL, Redis]
specializations: [Knowledge Graphs (Apache AGE, FalkorDB), CodeQL, Computer Vision, Triton]

Pinned Loading

  1. InfiniteContext-1B InfiniteContext-1B Public

    End-to-End MLSys Platform: From Distributed Training (SLURM) to SOTA Inference (Kubernetes)

    Python

  2. Document-Classification-using-LayoutLM Document-Classification-using-LayoutLM Public

    This PyTorch implementation of LayoutLM paper by Microsoft demonstrate the SequenceClassfication task using HuggingFaceTransformers to classify types of Documents.

    Jupyter Notebook 35 13

  3. US-Driver-License-data-extraction US-Driver-License-data-extraction Public

    Multi-approach ML pipeline for driver license data extraction using YOLO and OCR

    Jupyter Notebook 26 12

  4. ABP-estimation-using-PPG ABP-estimation-using-PPG Public

    PyTorch implementation of Estimating Arterial blood pressure (ABP) using Photoplethysmogram (PPG) signals.

    Jupyter Notebook 10

  5. Sleep-Awake-Classifiation-using-PPG Sleep-Awake-Classifiation-using-PPG Public

    PyTorch implementation of sleep-stage classifier using photoplethysmography (PPG) data.

    Jupyter Notebook 16 3

  6. information-retrieval information-retrieval Public

    RAG pipeline for AWS case studies: vector search, rerank, synthetic QA, fine-tuned LLaMA 3 8B. ChromaDB, sentence-transformers, Unsloth/LoRA.

    Jupyter Notebook 1