"Designing intelligence — from autonomous drone swarms to generative models to the quantum edge."

📍 Pune, India  |  🛰️ AI Research Intern @ DIAT-DU (DRDO)  |  🎓 B.Tech CSE (AI/ML) · CGPA 9.1  |  📄 2× Published  |  🏆 4× Awards

🧬   W H O   A M   I

class LakshitSaini:
    """
    Final-year AI/ML Engineer who doesn't just study intelligence —
    he builds it, publishes it, and ships it to production.
    Currently researching autonomous drone swarm AI at DRDO,
    while extending the stack into the quantum regime.
    """

    identity   = "AI Engineer × Researcher × Builder"
    university = "DY Patil International University, Pune"
    cgpa       = 9.1 / 10
    right_now  = "AI Research Intern @ DIAT-DU (DRDO) — defence swarm intelligence"

    current_obsessions = [
        "Multi-Agent RL (MAPPO) for 50-agent autonomous drone swarms",
        "Hybrid LSTM–Diffusion architectures for human-like synthesis",
        "Agentic LLM systems with multi-agent orchestration",
        "Quantum ML: variational circuits for hybrid quantum-classical models",
        "Speech disfluency detection via self-supervised learning",
        "AI security: prompt injection, Zero Trust & adversarial robustness",
    ]

    research_mission = """
        Bridge cutting-edge ML research and real-world production systems —
        from contested defence environments to the quantum edge.
    """

    superpower = "Turning research ideas into publication-grade papers AND deployed systems"
    fun_fact   = "Built a trajectory model so human-like it fooled 100% of bot detectors 🤫"

2 Publications NCTAAI 4.0 · IJSREM +1 under submission (IEEE)

3 Internships DRDO · Dexpert · ICAR Defence AI · SDE · Research

4 Awards National · State · University

6 Projects Research + Production 212,568-sample custom dataset

💼   E X P E R I E N C E

From national defence research to production-grade engineering — a harder problem every year.

🛰️	AI Research Intern · DIAT-DU (DRDO) — Pune, India  ·  May 2026 – Present Researching AI-driven autonomous drone swarm systems for defence applications — decentralized multi-agent coordination for cooperative search and target detection across swarms of up to 50 agents in 3D simulation. Developing MAPPO multi-agent reinforcement learning policies, YOLOv8-based aerial target detection, Zero Trust security architecture, and adversarial AI robustness frameworks for swarm intelligence in contested defence environments.
⚙️	Software Engineering Intern · Dexpert Systems Pvt. Ltd. — Pune, India  ·  May 2025 – July 2025 Engineered production-grade RESTful microservices with Node.js, Express.js, TypeScript, Sequelize ORM, and MySQL — modular, independently deployable architecture serving enterprise client modules. Integrated Razorpay payment APIs and CryptoJS-encrypted transaction pipelines with OTP-based authentication, delivering secure end-to-end financial backend workflows.
🌾	Research Intern · ICAR-NIASM — Baramati, India  ·  May 2024 – July 2024 Leveraged DelPro dairy-processing software to organize and analyze farm-level operational data, enabling data-driven decisions for livestock management and productivity. Designed structured data-collection templates and conducted field interviews with farmers to standardize agricultural data workflows.

⚛️   Q U A N T U M   F R O N T I E R

Where classical ML plateaus, I start asking quantum questions.

My newest research direction: pushing machine learning beyond classical limits. I'm building intuition and prototypes around hybrid quantum–classical models and privacy-preserving distributed learning — the natural next layer on top of my swarm, diffusion, agentic, and behavioral-AI work.

⚛️ Quantum Machine Learning Variational Quantum Circuits (VQC) as trainable layers Quantum kernels & feature maps for classification Hybrid quantum–classical autograd pipelines Barren-plateau-aware ansatz design

🔐 Quantum Federated Learning Distributed QML across non-IID clients Parameter aggregation of quantum circuit weights Privacy-preserving training without raw-data sharing Bridging my Zero-Trust & LLM-security work into quantum trust models

🌌 The Convergence Map

How every project I've built feeds the quantum trajectory — rendered live by GitHub:

%%{init: {'theme':'dark','themeVariables':{'primaryColor':'#0d1b2a','primaryTextColor':'#00d4ff','primaryBorderColor':'#7c3aed','lineColor':'#f5a623','fontSize':'14px'}}}%%
graph LR
    A[🧠 MIMIC<br/>hybrid LSTM × Diffusion] -- "hybrid architecture DNA" --> Q{{⚛️ Hybrid<br/>Quantum-Classical<br/>Models}}
    G[🛰️ DRDO Drone Swarms<br/>MAPPO × YOLOv8] -- "decentralized coordination & trust" --> F{{🔐 Quantum<br/>Federated<br/>Learning}}
    B[🛡️ AI Security Taxonomy<br/>NCTAAI 4.0 🏆] -- "adversarial & trust research" --> F
    C[📗 Edge ML on MCUs<br/>IJSREM] -- "resource-constrained learning" --> F
    D[🤖 InsightCrew + StonkSensei<br/>multi-agent systems] -- "distributed orchestration" --> F
    E[🗣️ Speech Disfluency<br/>self-supervised Wav2Vec 2.0] -- "representation learning" --> Q
    Q --> Z((🚀 Quantum AI<br/>Research<br/>2026 →))
    F --> Z

    style Q fill:#0d1b2a,stroke:#00d4ff,stroke-width:2.5px
    style F fill:#0d1b2a,stroke:#a78bfa,stroke-width:2.5px
    style Z fill:#0d1b2a,stroke:#f5a623,stroke-width:3px

Loading

_{🟣 Active learning & prototyping track — not yet published. Open to research collaboration in this space.}

⚔️   T E C H   A R S E N A L

🧠   AI / ML / Deep Learning

🛰️   Autonomous Systems & Reinforcement Learning

🤖   LLM & Agentic AI

⚛️   Quantum & Emerging

💻   Languages, APIs & Infrastructure

🎯   Computer Vision & Signal Processing

🚀   F E A T U R E D   P R O J E C T S

👆 Click any project to expand the details.

🧠   MIMIC — Hybrid LSTM–Diffusion Trajectory Synthesis

The research that didn't just pass academic review — it fooled bot detectors.

Problem: Existing motion synthesis models are either too rigid (LSTM) or too noisy (pure diffusion) to produce convincingly human-like behavioral trajectories. Innovation: A hybrid architecture where an LSTM encodes contextual priors that condition a residual DDPM diffusion model — combining sequential memory with stochastic refinement. Results: 📊 JSD = 0.1478 (near-perfect human distribution match) 📈 Smoothness Index = 0.8593 🛡️ Bot-Detector Evasion = 100% (vs 0% LSTM-only baseline) 📦 Custom dataset: 212,568 samples across 4 behavioral profiles ⚙️ Benchmarked ADE/MSE + inference latency across evaluation conditions

Status: 📝 IEEE Submission In Progress Behavioral AI Trajectory Synthesis Bot Evasion Generative Models

🔍   InsightCrew — Multi-Agent Research & Content Engine

Three agents. One goal. Zero manual research.

Problem: Research synthesis and content generation require hours of parallel, context-aware work — impossible for a single LLM call. Innovation: A sequential 3-agent CrewAI pipeline (Researcher → Analyst → Writer) powered by Gemini 2.0 + Tavily RAG for real-time web context injection. Impact: ♾️ Fully autonomous research-to-content pipeline 🌐 Production-deployed on Streamlit Cloud — 100% Python ⚡ Real-time RAG via Tavily API

Agentic AI Multi-Agent RAG LLM

📈   StonkSensei — Agentic LLM Financial Intelligence System

Real-time financial intelligence, dual-agent architecture, streaming inference.

Problem: Financial analysis requires aggregating real-time data, news, and quantitative reasoning simultaneously — a perfect multi-agent problem. Innovation: Dual-agent phidata system with Groq's deepseek-r1-distill-llama-70b for streaming inference, DuckDuckGo + Yahoo Finance as live data tools, served via FastAPI + Uvicorn. Impact: ⚡ Low-latency streaming LLM responses 📡 Live market + news data integration 🖥️ Interactive agent playground UI

Fintech AI Streaming LLM Agentic

🗣️   Speech Disfluency Detection & Remediation

Fine-tuning self-supervised speech models for clinical-grade stutter detection.

Problem: Automated disfluency detection for real-world accessibility systems requires robustness across noise conditions and accurate classification of subtle speech patterns. Innovation: Fine-tuned Wav2Vec 2.0 on SEP-28k for 5-class disfluency detection + a Whisper-aligned remediation pipeline with acoustic robustness validation. Results: 🎯 WER < 0.15 (Whisper-aligned transcription) 🔊 Validated across 5–25 dB SNR noise conditions 📐 Evaluated MOS + LPC metrics (LLR, ISD) for remediation quality 🏥 5 stutter categories classified

Status: 🔬 Active Research · Jan 2026 – Present Speech ML Accessibility Self-Supervised

🤟   SignSpeakAI — Real-Time Sign–Speech Translation

Making communication barrier-free with multimodal AI.

Innovation: Bidirectional multimodal pipeline — MediaPipe + LSTM for Sign-to-Speech, and a Speech-to-Sign TTS module — with multilingual support and modular CV/NLP architecture. Impact: Real-time, low-latency accessible HCI interaction across multiple languages.

HCI Accessibility Multimodal AI

🛰️   INR-AD — Anomaly Detection via Implicit Neural Representations

SIREN networks meet satellite telemetry.

Unsupervised anomaly scoring on NASA SMAP satellite telemetry using a SIREN architecture (implicit neural representations). Fully modular, reproducible codebase with benchmarked results.

SIREN Anomaly Detection Satellite Telemetry Unsupervised Learning

📄   R E S E A R C H   &   P U B L I C A T I O N S

📘	Exploring the Landscape of AI Security: A Taxonomy Review on Prompt Injection and Jailbreaking NCTAAI 4.0 National Conference Proceedings · May 2025 A systematic taxonomy of prompt injection and jailbreaking attack mechanisms across large language models, analyzing adversarial implications for AI safety in Industry 4.0 contexts — cybersecurity and autonomous systems.
📗	Enhancing Predictive Maintenance in Smart Home Devices using AI/ML on Microcontrollers International Journal of Scientific Research in Engineering and Management (IJSREM) · October 2024 On-device ML deployment for real-time fault detection in resource-constrained IoT microcontrollers, reducing cloud inference dependency with edge AI methodology.
📙	MIMIC: Hybrid LSTM–Diffusion Model for Human-Like Trajectory Synthesis (Under Submission) Hybrid LSTM prior + conditional residual DDPM for context-aware behavioral trajectory generation. JSD = 0.1478, 100% bot-evasion rate. Targeting IEEE Transactions on Biometrics, Behavior and Identity Science.

Research Interests

Multi-Agent RL & Swarm Intelligence · LLM Security & Adversarial AI · Generative Models · Speech ML Behavioral Biometrics · Agentic Systems · Quantum ML & Quantum FL · Edge AI / TinyML · Implicit Neural Representations

🏆   A C H I E V E M E N T S   W A L L

🥇	Achievement	Issuer	Year
🎓	Academic Excellence Scholarship	DY Patil International University	2026
🏆	1st Rank — Ideathon 2025	DY Patil International University	Nov 2025
📜	Research Paper Award	NCTAAI 4.0 · ICEM National Conference	Mar 2025
🥇	1st Rank — State-Level Research Paper Competition	MITACSC	Dec 2024

📊   G I T H U B   A N A L Y T I C S

⚡   Recent Activity  _{(auto-updates daily via GitHub Actions — requires activity.yml workflow, provided)}

1. 💪 Opened PR #24 in YeshitaMotwani/AutoStartup-Simulator
2. 💪 Opened PR #23 in YeshitaMotwani/AutoStartup-Simulator
3. 💪 Opened PR #2 in Faiza-Bagban/AutoStartup-Simulator
4. 🎉 Merged PR #21 in YeshitaMotwani/AutoStartup-Simulator
5. 💪 Opened PR #21 in YeshitaMotwani/AutoStartup-Simulator

🕰️   C A R E E R   T I M E L I N E

2023 ──────────────────────────────────────────────────────────────── 2027
  │
  ├─── Aug 2023 ──▶  🎓  B.Tech CSE (AI/ML) · DY Patil International University
  ├─── May 2024 ──▶  🌾  Research Intern · ICAR-NIASM · Agricultural Data Systems
  ├─── Oct 2024 ──▶  📗  Published: Predictive Maintenance on Microcontrollers · IJSREM
  ├─── Dec 2024 ──▶  🥇  1st Rank · State-Level Research Paper · MITACSC
  ├─── Feb–May 2025 ▶  🤟  SignSpeakAI · Multimodal Sign–Speech Translation
  ├─── Mar 2025 ──▶  📘  Research Paper Award · NCTAAI 4.0 · ICEM
  ├─── May 2025 ──▶  💼  Software Engineering Intern · Dexpert Systems Pvt. Ltd.
  ├─── May 2025 ──▶  🤖  InsightCrew · Multi-Agent Research Engine (Production)
  ├─── May 2025 ──▶  📈  StonkSensei · Agentic Financial Intelligence System
  ├─── Nov 2025 ──▶  🧠  MIMIC Research Started · Hybrid LSTM–Diffusion
  ├─── Nov 2025 ──▶  🏆  1st Rank Ideathon 2025 · DY Patil International University
  ├─── Jan 2026 ──▶  ✅  MIMIC Complete · JSD=0.1478 · 100% Bot Evasion
  ├─── Jan 2026 ──▶  🗣️  Speech Disfluency Detection Research · Active
  ├─── 2026 ─────▶  🎓  Academic Excellence Scholarship · DYPIU
  ├─── 2026 ─────▶  ⚛️  Quantum ML & Quantum FL · Exploration Track Begun
  ├─── May 2026 ──▶  🛰️  AI Research Intern · DIAT-DU (DRDO) · Autonomous Drone Swarm AI
  └─── 2027 ─────▶  🚀  B.Tech Graduation · Ready to build the future

🗺️ Flight Plan · 2026 → 2030

%%{init: {'theme':'dark','themeVariables':{'primaryColor':'#0d1b2a','primaryTextColor':'#00d4ff','lineColor':'#f5a623','fontSize':'13px'}}}%%
timeline
    title From Classical Orbit to Quantum Deep Space
    2026 : AI Research Intern @ DIAT-DU (DRDO) — Drone Swarm AI : Ship Speech Disfluency System : MIMIC → IEEE submission : First VQC prototypes (PennyLane · Qiskit)
    2027 : B.Tech graduation : Hybrid quantum-classical paper : Open-source QML tooling
    2028 : Graduate research — Quantum ML : QFL simulation framework
    2029 : Publish in QML venues : Quantum NLP / CV experiments
    2030 : Lead an AI × Quantum research initiative

Loading

🎯   C U R R E N T   F O C U S

🚀 Currently Building Autonomous drone swarm AI @ DRDO — MAPPO policies for 50-agent coordination YOLOv8 aerial target detection + Zero Trust swarm security Speech Disfluency Detection with real-time remediation MIMIC IEEE paper finalization & journal submission 📚 Researching Decentralized multi-agent coordination in contested environments Adversarial AI robustness for defence-grade systems Diffusion architectures for controllable generation LLM safety: adversarial prompt attacks & defenses

🎯 Currently Learning Quantum ML & Quantum Federated Learning Advanced multi-agent RL & swarm orchestration patterns First VQC prototypes (PennyLane · Qiskit) 🤝 Open to Collaborate On ML Research (publication-grade) Agentic AI / LLM Systems & Quantum ML Autonomous Systems, Defence AI & Security AI for Healthcare & Accessibility Hackathons & Research Internships

🧬   D E V E L O P E R   D N A

A machine-learning model trained on curiosity and caffeine.

Model Architecture: Lakshit-v4.0-Researcher-Builder

Training Data:
  - 3 internships — DRDO Defence AI · Production SDE · Applied Research
  - 3 years of AI/ML coursework & self-study
  - 212,568 custom behavioral motion samples (self-collected)
  - 2 published research papers (+1 under IEEE submission)
  - 6 production/research projects shipped

Fine-Tuning (current epoch):
  - Multi-Agent RL on 50-agent autonomous drone swarms @ DRDO

Activation Functions:
  - Curiosity-driven exploration
  - Research-first problem solving
  - Production-grade output standards

Loss Function:   Cross-entropy between "what exists" and "what should exist"
Optimizer:       Gradient descent on impact
Regularization:  Academic rigor + real-world validation
Next Layer:      Quantum-classical hybrid (training...)

Status: 🟢 TRAINING IN PROGRESS — Expected convergence: Aug 2027

📡   L E T ' S   B U I L D   S O M E T H I N G

Whether you're a researcher, recruiter, founder, or fellow builder — if you're working on something that matters, I want to talk.

Currently open to:

AI/ML & Research Roles   Research Internships   Paper Co-authorship   Defence & Autonomous Systems   Quantum ML Collabs   Hackathon Teams   Grad School

        |ψ⟩ = α|build⟩ + β|research⟩        — collapse the wavefunction: say hi.