Skip to content
View loki52501's full-sized avatar
2 followers · 12 following

Block or report loki52501

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
loki52501/README.md

⚡ whoami

class Lokesh:
    title      = "AI Engineer · Agentic Systems Designer · AI Solutions Architect"
    education  = [
        "M.S. Cybersecurity Operations & Technology — Penn State (3.7 GPA, May 2026)",
        "Integrated M.S. Information Technology — Anna University (3.3 GPA)",
    ]
    building   = [
        "Multi-agent RAG pipelines on Kubernetes",
        "LLM evaluation pipelines for sycophancy, CoT faithfulness & adversarial robustness",
        "AI-native security systems — threat detection, anomaly detection, IDS",
        "Browser-native AI tools with spaced-repetition & adaptive scheduling",
    ]
    stack      = ["Python", "PyTorch", "HuggingFace", "LangChain", "FastAPI",
                  "Docker", "Kubernetes", "AWS", "RAG", "ChromaDB", "Ollama"]
    certs      = ["Cisco CCNA", "CompTIA Security+"]
    philosophy = "Agents that reason well must also reason honestly."
    status     = "OPT-ready June 2026 · Open to relocation · Targeting AI infra & agentic roles"

"How do you know when a system's stated behavior reflects its actual internal state?"

This question connects everything I build — from IDS pipelines that detect hidden attack patterns, to LLM evaluators that catch post-hoc rationalization, to agentic systems that need auditable decision trails.

🤖 Agentic AI Systems

Projects where I architect, orchestrate, and ship multi-agent and LLM-powered systems end-to-end.

🐾 OpenClaw — Multi-Agent RAG on K8s

Node.js · Kubernetes · RAG · ChromaDB · CI/CD

Production multi-agent RAG system deployed on Kubernetes:

  • Agent orchestration layer with retrieval, synthesis, and response agents
  • ChromaDB vector store with semantic chunking pipeline
  • Full CI/CD — GitHub Actions → Docker → K8s rolling deploys
  • Designed for horizontal scale: agent pods scale independently on load

📚 SQ3R AI Study Agent — Chrome Extension

TypeScript · React · Node.js · SQLite · FSRS-4.5 · Multi-LLM

Browser-native AI agent transforming any page or PDF into a full study system:

  • Multi-LLM routing: local Ollama or cloud (Claude, OpenAI, Gemini, DeepSeek)
  • FSRS-4.5 spaced-repetition with dynamic stability & difficulty curves
  • Finite-state review machine with mutex-locked batch queue across service workers
  • Pomodoro interrupt agent auto-surfaces due cards without opening the extension

🧠 LLM Sycophancy + CoT Faithfulness Suite

Llama-3.2-3B / 3-8B · PyTorch · BIG-Bench-Hard · Streamlit

Two-part LLM evaluation suite probing internal alignment:

  • Sycophancy harness: agreement bias 45%, pressure capitulation 40%, flattery resistance 80%
  • CoT faithfulness: biasing-feature perturbation on BIG-Bench-Hard — does the chain of thought verbalize or silently conform?
  • Extends Turpin et al. (2023) & Lanham et al. (2023) on Llama-3-8B
  • Full harness + Streamlit dashboard + 13 unit tests on GitHub

🏗️ Agentic System Architecture — How I Design AI Solutions

╔══════════════════════════════════════════════════════════════════════════════╗
║                       AGENTIC AI SOLUTION BLUEPRINT                         ║
╠══════════════════════════════════════════════════════════════════════════════╣
║                                                                              ║
║   USER / APP  ──►  [ API Gateway / Prompt Interface ]                        ║
║                              │                                               ║
║                     ┌────────▼────────┐                                      ║
║                     │  Planner Agent  │  ← Task decomposition                ║
║                     │  (Orchestrator) │  ← Tool selection                    ║
║                     └────────┬────────┘  ← Confidence gating                ║
║                              │                                               ║
║          ┌───────────────────┼───────────────────┐                          ║
║          ▼                   ▼                   ▼                          ║
║   [ Retrieval Agent ]  [ Synthesis Agent ]  [ Action Agent ]                ║
║     ChromaDB / RAG      LLM reasoning         Tool APIs                     ║
║          │                   │                   │                          ║
║          └───────────────────┼───────────────────┘                          ║
║                              │                                               ║
║                     ┌────────▼────────┐                                      ║
║                     │  Eval / Monitor │  ← Sycophancy & CoT faithfulness     ║
║                     │     Layer       │  ← Behavioral consistency checks     ║
║                     │  (LLM harness)  │  ← Confidence-based escalation       ║
║                     └────────┬────────┘                                      ║
║                              │                                               ║
║                     [ Audit Log / Response ]  ──►  USER                      ║
║                                                                              ║
╚══════════════════════════════════════════════════════════════════════════════╝

🛡️ AI Security & Threat Detection

🌊 DDoS Detection IDS (Capstone)

BiLSTM · 1D Residual CNN · PyTorch · BCCC-Cloud-DDoS-2024

Ensemble IDS on 700K+ network flows:

  • 94.11% accuracy — Attack / Benign / Suspicious (3-class)
  • 318 features → 86 via selection; sliding temporal windows
  • Confidence escalation auto-classifies 85% of traffic, routes uncertain flows for review
  • Architecture lineage: UNSW-NB15 research → refined for cloud DDoS

🔬 Network IDS Research — Anna University

LSTM · 1D CNN · SMOTE · UNSW-NB15

Foundational deep-learning IDS pipeline:

  • LSTM + 1D CNN with SMOTE for minority attack class balance
  • Temporal window analysis directly seeded DDoS capstone architecture
  • AI Research Intern, Anna University (2023–2024)

🔧 Engineering & Platform Projects

Project Stack What It Does
🎬 Video Streaming Platform MinIO, HEVC, FFmpeg Adaptive HEVC transcoding, object-storage backend
📚 E-Library React, React Native, FastAPI Cross-platform library management (web + mobile)
🩸 Blood Donor App Spring Boot, PostgreSQL Microservice donor matching + notification system
⚡ Static Site Generator C++ Zero-dependency SSG from scratch
🌐 QUIC Router Simulation Python Protocol-level simulation of QUIC routing
🌾 Agro Foods Platform React, FastAPI, ML Produce grading via image classification + billing automation

💼 Experience


🎓

Instructional Assistant — Data Structures & Algorithms
Pennsylvania State University · Aug 2024 – Present
Teaching Java & Haskell to 100+ graduate students across DSA and Discrete Mathematics. Weekly instruction and office hours on algorithmic correctness and asymptotic complexity.
🔬

AI Research Intern — Network Intrusion Detection
Anna University, College of Engineering Guindy · Feb 2023 – Jul 2024
Deep-learning IDS pipeline on UNSW-NB15 (LSTM + 1D CNN + SMOTE). Temporal window analysis directly seeded the 94.11% accuracy DDoS capstone architecture.
☁️

Platform Engineer Intern
Tvastr
AWS compliance automation engine spanning 1,000+ accounts. Elasticsearch query optimization for infrastructure observability at scale.
🌾

ML Software Engineer
Lokesh Agro Foods
Image classification for produce quality grading. Automated billing and inventory systems for family-owned organic food business.

🛠️ Tech Stack

Languages

AI / LLM / Agentic Systems

HuggingFace LangChain RAG ChromaDB Ollama OpenAI Anthropic Multi-Agent FSRS

Backend & APIs

Frontend

Cloud & Infrastructure

Lambda SQS DynamoDB CI/CD

Security & Networking

Wireshark Snort TCP/IP · QUIC · SIEM · Nmap · Elasticsearch

📊 GitHub Stats

🏅 Certifications

  

Snake animation

🚀 Open to Roles In

AI Engineer Agentic Systems Designer AI Solutions Architect LLM Infra Engineer AI Security Engineer


OPT-ready ~June 2026 · Open to relocation · US work authorized




Agents that reason well must also reason honestly. ☕

Pinned Loading

  1. autoencode_detection_research-Project autoencode_detection_research-Project Public

    Jupyter Notebook 1

  2. bloodbank-webapp bloodbank-webapp Public

    1

  3. MCP_cloudsec MCP_cloudsec Public

    Python 1

  4. nim-chat-app nim-chat-app Public

    Nim 1

  5. notehost notehost Public

    Forked from velsa/notehost

    Free Hosting for Notion Sites!

    TypeScript 1

  6. whatsappbot whatsappbot Public

    JavaScript 1