Skip to content
View sibi-seeni's full-sized avatar
1 follower · 0 following

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull Sharkx2

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull Sharkx2

Block or report sibi-seeni

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. 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
sibi-seeni/README.md

Hi, I'm Sibi! 👋🏼

Typing intro

AI and data engineer with client-facing delivery across enterprise consulting, financial services, defense, and regulated medtech.
I've taken a legacy warehouse to a Spark lakehouse through go-live, and now I'm focused on AI security:
making agentic and RAG systems robust enough for high-stakes deployments.

Right Now

  • AI Security Engineer with NIWC & DIU through Hacking for Defense, reproducing RAG poisoning attacks from 8 papers and benchmarking defenses for a Navy-facing prototype
  • Researching trust-aware graph retrieval, a GNN retriever that down-weights poisoned sources before they reach the LLM (extending into my Spring 2027 capstone)
  • Under review at ECJ Data Science (Elsevier): modeling how fast community backlash decays online
  • Open to AI Security, Agentic AI, and Data Engineering roles · graduating May 2027

Experience

AI Security Engineer · Naval Information Warfare Center (NIWC) & DIU, Hacking for Defense · Aug 2026 – Present
  • Scoped the threat through 19 stakeholder interviews (Navy captains, NIWC directors, portfolio managers, engineers), narrowing 3 classes of RAG poisoning down to the highest-impact one: outdated or conflicting document versions that mislead even agentic RAG
  • Reproduced attacks from 8 published papers to set evaluation baselines; now benchmarking defenses for a planned prototype
  • Translating technical risk into defense-in-depth recommendations for non-technical DoD stakeholders
Software Engineer Intern · Verotech Solutions LLC · May 2026 – Aug 2026
  • Built an agentic RAG decision-support system for Stryker, turning FDA 21 CFR 820 requirements into a client-facing compliance workflow
  • Added source attribution and human-in-the-loop review using Google's Agent2Agent (A2A) protocol, wired into the client's existing compliance system
  • Scaled reasoning to 200+ pages of compliance docs across 6-turn sessions, owning the Python codebase through review and handoff
Technology Consulting Associate · PwC India · Aug 2024 – Jul 2025
  • Forward-deployed for 6 months with Aditya Birla Sun Life, working with the Head of Data inside a 40-member Agile team
  • Owned 50+ Teradata → Databricks DDL conversions and 40+ silver/gold DML scripts for a medallion lakehouse fed by Azure Data Factory
  • Ran daily validation with 6 business units through go-live, letting the client decommission Teradata and cut cloud costs 17%
  • Built the client's CI/CD with automated testing (30% faster deploys) and Power BI observability dashboards (3× faster insights)
  • Built LLM proofs-of-concept presented in sales engagements that helped land a new enterprise project
Graduate Teaching & Research Assistant · University of Florida (IC3 Lab; LATTE Lab - Engineering Education) · Nov 2025 – Present
  • Built an agent-based NL2SQL tool (LangChain) giving an LLM function-calling access to the MIMIC critical-care database
  • Improved transcription accuracy 33% with a locally deployed GPT-OSS-20B
  • Maintain server infrastructure and CI for 600+ students a semester; built analytics pipelines presented at the UF Engineering Education Symposium

Featured Work

Xposure: Accessible Portfolio Exposure Graph

Honorable Mention, Blackstone challenge @ ShellHacks 2026 (solo, 36 hrs)

Shows how "diversified" holdings are secretly linked, like utilities and semiconductors through data-center power demand. A PostgreSQL knowledge graph with 20+ FastAPI endpoints for multi-hop traversal and shock propagation, plus a Snowflake Cortex agent that proposes new relationships for human approval. Built for blind and low-vision investors with narration and sonification.

FastAPI PostgreSQL Snowflake Cortex React TypeScript PyTest

Trust-Aware Graph Retrieval (in progress)

UF Applied ML II architecture project → Spring 2027 capstone

A GNN retriever with attention that models documents and evidence as a graph, learning to down-weight poisoned or untrustworthy sources before they reach the LLM in RAG or agent-memory pipelines. Complements the adversarial RAG work with NIWC.

PyTorch GNNs RAG AI Security

The Half-Life of Outrage

Submitted to ECJ Data Science (Elsevier)

Aspect-based sentiment pipeline over 160k+ YouTube comments across 905 videos using LLaMA 3.3 70B and RoBERTa (Gwet's AC1 of 0.98–0.99 vs. gold labels). Exponential decay models show AI-voice controversies fade 4× faster (0.48 days) than general game discourse.

LLaMA 3.3 RoBERTa NLP Statistical Modeling

Educator's AI Analytics Dashboard

GatorHacks 2025

Upload a lecture, get a transcript, topics covered, and progress against the syllabus. FastAPI backend chaining Whisper v3 → Llama-3.3-70B, with async background jobs cutting latency 40%.

FastAPI Whisper Llama 3.3 TypeScript asyncio

More Builds

Real-Time Product Launch Radar
Event-driven pipeline on the live Bluesky firehose with INT8 ONNX sentiment inference, ClickHouse time-series storage, and Grafana dashboards tracking sentiment velocity.
ONNX ClickHouse Grafana Streaming

TSLA Options on Black-Scholes
21k+ contracts, 11 features. XGBoost cut MAPE 46% vs. regression and beat a from-scratch Bi-LSTM by 1.8%.
XGBoost PyTorch Time Series

Tech Toolbox

AI & Agents

Data & Cloud

Languages & Engineering

Security Testing

Education & Recognition
University of Florida · MS, Applied Data Science · May 2027
NIT Trichy · B.Tech, Electrical & Electronics Engineering · May 2024
  • UF Herbert Wertheim College of Engineering Achievement Award
  • Microsoft Certified: Power BI Data Analyst Associate · Python & Pandas for Data Engineering (Duke) · Orchestrating GenAI Workflows (DeepLearning.AI)

Always up for research collaborations, AI security conversations, or data-heavy product ideas.
Reach me at sibiseenivasan@gmail.com or on LinkedIn.

Pinned Loading

  1. realtime-product-radar realtime-product-radar Public

    An end-to-end, event-driven streaming pipeline that captures live social commentary from the Bluesky Jetstream firehose, runs low-latency INT8 ONNX sentiment inference, stores time-series aggregati…

    Python

  2. half-life-of-outrage half-life-of-outrage Public

    The Half-Life of Outrage: Modeling Temporal Sentiment Decay and Community Backlash in Video Game Announcements

    Python

  3. professors-dash professors-dash Public

    AI-powered backend service designed to provide university professors with deep, actionable insights into their lectures. Uses LLMs like OpenAI Whisper v3 and Llama 3.3 to transcribe, analyze and st…

    Python

  4. accessible-portfolio-graph accessible-portfolio-graph Public

    Xposure, an accessible audio-visual knowledge graph that reveals hidden economic overlaps across public equity, private equity, real estate, private credit and infrastructure holdings.

    Python

  5. qwen3-local-mlx qwen3-local-mlx Public

    Run the Qwen3 language model locally on Apple Silicon (M4/M3/M2/M1) using MLX. This project shows how to quantize, load, and chat with Qwen3-0.6B efficiently on macOS — all offline.

    Python 1

  6. credit-churn-deploy credit-churn-deploy Public

    High-performance Credit Card Churn prediction service informed by financial industry experience. Predicts customer attrition using an optimized XGBoost/LightGBM model for high accuracy. Full MLOps …

    Jupyter Notebook