🚀 I'm a CS sophomore at VIT Amaravati specializing in AI & ML, building distributed AI systems, LangGraph-orchestrated multi-agent pipelines, and production-grade backends. Actively researching EEG-to-text brain-computer interfaces with OpenBCI hardware, exploring commercialization pathways in assistive tech and neurorehabilitation. From satellite-validated flood prediction systems to geolocation comedy platforms to ESP32 smart glasses — I build end-to-end, from silicon to software to screen.
|
🏅 VIT Internal Expo Smart Vision Aid Selected |
🎓 CGPA 8.22 B.Tech CS · AI & ML |
🔬 Active BCI Research EEG → Text + Robotics |
🌊 Flood-Guard AI Satellite-Validated Simulation |
🤖 13+ Projects Shipped & Deployed |
🎬 50+ Edits Editor Cyclops · 3+ Yrs |
class AnkitSengupta:
def __init__(self):
self.name = "Ankit Sengupta"
self.role = ["AI Engineer", "Backend Developer", "BCI Researcher", "Embedded IoT Dev"]
self.university = "VIT Amaravati · Class of 2028"
self.cgpa = 8.22
self.skills = {
"AI / ML" : ["RAG Pipelines", "AI Agents", "LangGraph", "LangChain",
"TensorFlow", "YOLO", "OpenCV", "XGBoost", "RL Feedback"],
"LLM Infra" : ["LangSmith", "ChromaDB", "FAISS", "Qdrant", "HuggingFace",
"Gemma-4", "MedGemma-4B", "Phi-3", "Prompt Engineering"],
"Research" : ["OpenBCI", "EEG Signal Processing", "MNE-Python",
"Hydrodynamic Simulation", "Google Earth Engine", "BCI Robotics"],
"Systems" : ["Distributed Systems", "Docker", "Kubernetes",
"Self-Healing Architecture", "Fault Tolerance"],
"Backend" : ["Django", "FastAPI", "Flask", "Node.js",
"MongoDB", "PostgreSQL", "Neo4j", "Redis"],
"Frontend" : ["React", "Next.js", "TypeScript", "Vite", "Framer Motion"],
"Languages" : ["Python", "C", "C++", "JavaScript", "TypeScript", "Java", "Dart"],
"Mobile" : ["Flutter", "Android", "Kotlin", "Expo", "React Native", "Firebase"],
"Embedded" : ["ESP32", "Arduino", "C++", "IoT Sensors"],
"Video Editing": ["Premiere Pro", "After Effects", "DaVinci Resolve", "Motion Design"],
}
self.active_research = "🔬 EEG → Text BCI + Robotic Implementation (OpenBCI)"
self.status = "🟢 Open to internships & collaborations!"
def say_hi(self):
print("Let's build something that matters 🚀")Self-Employed · Remote · 2023 — Present
Independently run an end-to-end video editing and content strategy practice under the brand Editor Cyclops, serving 10+ creator-economy clients with short-form content for Instagram Reels, YouTube Shorts, and brand campaigns.
| 🎯 Production Pipeline | End-to-end editing for 10+ concurrent clients — brief to final delivery across vertical, horizontal, and square formats |
| ⚡ Signature Style | Fluent in Devin Jatho-style editing — rhythm-locked cuts, deliberate colour contrast, beat-synced motion, hook-first energy |
| 🍎 Motion & UI Design | Apple-style Glass UI / liquid-glass animations, kinetic typography, layered motion graphics for app promos |
| 📊 Data-Driven Strategy | Engagement data applied to pacing, hooks, and colour treatment for platform-native performance |
| 📈 Track Record | 50+ projects · 100% client satisfaction · Premiere Pro · After Effects · DaVinci Resolve |
🎬 View Portfolio → · 📸 @editorcyclops →
🔥 "Every project is a chance to bridge theory and real-world impact"
Actively researching novel approaches to decode raw EEG brainwave signals captured via OpenBCI hardware into text commands and robotic control inputs — with a long-term focus on commercialization in assistive technology, neurorehabilitation, and human-robot collaboration.
| 🧠 EEG Capture | OpenBCI hardware for non-invasive brainwave signal capture |
| 📊 Signal Processing | Multi-band analysis: alpha, beta, theta, delta wave decomposition (MNE-Python) |
| 🔤 EEG → Text Pipeline | Transformer-based decoder under active development — mapping neural patterns to text outputs |
| 🤖 Robotic Actuation | Investigating direct robotic control from decoded thought signals |
| 💡 New Strategy | Actively developing new architectural and signal-processing approaches to improve decode accuracy |
| 🏭 Commercialization | Targeting assistive technology, neurorehabilitation, and human-robot collaboration markets |
🔬 Actively researching alongside other projects — developing new strategies and real-world implementation approaches for this technology. This research area holds significant commercial potential and I am committed to advancing it toward production-grade deployment. View Repo →
A high-performance, modular, locally-hosted AGI OS — self-healing, horizontally scalable, and containerized — that reasons, remembers, monitors the world, teaches, and assists clinically, without sending a single byte to the cloud.
| 🧠 General Chat | Local Llama/Qwen models for offline reasoning |
| 🏥 Health Track | MedGemma 4B with SigLIP vision — X-ray & clinical reasoning; FHIR-compatible EHR; zero patient data exfiltration |
| 🎓 Education Track | Adaptive Tutor with OCR homework grading, dynamic quiz generation, culturally localized content |
| 🌍 World Monitor | Live GDELT + USGS feeds → AI-synthesized geopolitical & disaster briefings + instability scoring |
| 🕸️ Mirofish Graph | Force-directed semantic knowledge graph with Leiden community detection — 10,000+ nodes tested |
| 🎙️ Jarvis Voice | Local Whisper STT + Kokoro/ONNX TTS — full duplex with interrupt support |
| 🛠️ Self-Healing Ops | Automatic health monitoring, failure detection, and service restart across all distributed components |
| ⚡ Performance | 60+ tokens/sec (RTX 3070+) · MedGemma 4B ~6GB VRAM · Cold start < 8s |
🏆 Kaggle × Google DeepMind Gemma 4 Good Hackathon · Health, Education & Global Resilience · View Repo →
Given a dam, reservoir state, and breach scenario — it runs a full hydrodynamic simulation, predicts WHERE the flood goes AND WHEN it arrives, validates against satellite imagery, and converts everything into a time-aware evacuation plan with a ranked, auditable emergency action list.
| 🌊 Hydrodynamic Engine | SPH (breach zone) + Delft3D (grid propagation) — real terrain from SRTM DEMs via WebGL/Three.js 3D visualization |
| ⏱️ Arrival-Time Mapping | Time-to-flood bands (0–15min, 15–30min, 30–60min, 1–3hr) per location — not just flooded/not-flooded |
| 🛣️ Dynamic Route Safety | Every road/bridge segment evaluated as a function of time against depth/velocity — routes that are "safe now but unsafe in 20 min" are flagged |
| 🛰️ Satellite Validation | Sentinel/Landsat via Google Earth Engine — observed water extent vs. simulated (IoU, agreement %, over/under-prediction zones) |
| 🚨 Emergency Priority Engine | Ranked action list across all villages scored on flood probability + depth + velocity + arrival time + population + accessibility |
| 🔍 Explainability Trail | Every recommendation carries "Why is Village A Priority 1?" — inputs used, model/scenario, assumptions, confidence, validation score |
| 📶 Offline Mode | Pre-downloadable scenario data — field teams can view results & generate PDF reports without live connectivity |
AI-powered stock direction prediction using an 8-node distributed LangGraph agent pipeline — real-time news, LLM reasoning, RAG pipelines, concurrency & multi-threading, and a reinforcement learning feedback loop that retrains on real outcomes after 3 days.
| 🔗 Orchestration | 8-node distributed LangGraph pipeline: fetch → sentiment → embed → stock → RAG → LLM → ML → ensemble, parallelized with concurrency & multi-threading |
| 🧠 LLM Core | Gemma-3-4B-IT (local HuggingFace) for news reasoning & directional prediction |
| 📦 RAG Layer | ChromaDB / FAISS vector store + sentence-transformers all-MiniLM-L6-v2 |
| 🔁 RL Feedback | Reward-weighted XGBoost — record → resolve → retrain on real outcomes after 3 days |
| 🔍 Observability | LangSmith tracing for full pipeline visibility |
An AI Teams co-pilot that turns meeting transcripts into structured, searchable knowledge artifacts — surfacing decisions and action items using Azure OpenAI GPT-4o, Semantic Kernel, Cosmos DB, and Azure Cognitive Search.
| 🎙️ Meeting Ingestion | Captures Teams meeting transcripts via Bot Framework → Azure Functions → Semantic Kernel agent pipeline |
| 🔍 Natural-Language Search | Azure Cognitive Search with vector + semantic indexing for conversational retrieval over past meetings |
| ✅ Action-Item Sync | Automatically surfaces decisions → syncs action items to Microsoft Planner via Graph API |
| 🔐 Enterprise Auth | MSAL SSO with Teams Tab + Azure App Service deployment |
A security OS for AI agent swarms — Prompt Injection Firewall, Neo4j Trust Graph, Self-Healing Daemon with snapshot/rollback, and a built-in Attack Simulator to validate swarm resilience against prompt injections and context overflows.
| 🔥 Prompt Injection Firewall | Real-time evaluation of agent inputs against heuristic + LLM-based injection patterns |
| 🕸️ Neo4j Trust Graph | Dynamic trust modeling — agents earn/lose trust based on output validity; below-threshold agents sandboxed |
| 🛠️ Self-Healing Daemon | Automated snapshotting + rollback of compromised agents using isolated memory boundaries (Redis) |
| 🧪 Attack Simulator | Built-in endpoints to safely simulate API poisonings, context overflows, and prompt injections |
| 🔄 Hot-Swappable LLMs | Instantly switch between OpenAI, Azure, and local LM Studio models mid-execution via UI |
A Turborepo monorepo connecting open-mic performers with their ideal audience via geolocation-based video matching — think Tinder for comedy. Web + iOS + Android + API in one codebase.
-000020?style=for-the-badge&logo=expo&logoColor=white){kind=icon}
| 🏗️ Architecture | Turborepo monorepo: apps/web (Next.js 14) + apps/mobile (Expo 51) + apps/api (Express) + shared packages |
| 📍 Geolocation Matching | GPS-based performer ↔ audience matching — swipe to discover local talent, attend live events |
| 🎥 Video Engine | Platform-agnostic sliding-window video memory manager (@comic-companion/video-engine) |
| 📦 Shared Types | @comic-companion/types — all TypeScript interfaces shared across web, mobile, and API |
The campus food delivery platform for VIT-AP University students — order from any canteen, track in real-time, straight to your hostel. Progressive Web App + Flutter native admin panel.
| ⚡ Real-Time Orders | Firestore onSnapshot for live order tracking across customer, restaurant, and admin roles |
| 📱 PWA | Installable on iOS & Android — works offline, responsive mobile-first glassmorphism UI |
| 🔔 Admin App | Flutter native Android admin panel with FCM push notifications on every new order |
| 📊 Live Analytics | Real-time revenue chart, top dishes leaderboard, order status pie from Firestore |
| 💬 WhatsApp Integration | One-tap customer chat from admin order view |
Real-time AI-powered assistive smart glasses with 3× synchronized ESP32-CAM modules, multi-threaded ultrasonic telemetry, and YOLO object detection for sub-100ms obstacle identification.
| 📷 Vision System | 3× synchronized ESP32-CAM modules + ultrasonic sensors → 360° spatial coverage |
| ⚡ AI Detection | YOLO real-time object detection — sub-100ms obstacle identification |
| 📱 Mobile App | Android companion (Java/Kotlin) with live audio alerts & navigation feedback |
| 🔧 Embedded Core | C++ on Arduino IDE — ultra-low-latency, concurrency-safe multi-threaded firmware |
| 🏅 Achievement | Selected for VIT Amaravati Internal Expo |
AI-powered platform connecting farmers and buyers through TensorFlow crop analysis, OpenCV disease detection, and Django REST marketplace.
Zero-effort expense tracking via secure Android bank SMS parsing — Firebase auth, real-time dashboard.
TensorFlow sentiment analysis + Pytesseract medical report parsing — emotion intelligence meets medical data.
Personalized AI health assistant for intelligent monitoring and conversational health insights.
╔══════════════════════════════════════════════════════════════════════════╗
║ 🧭 On My Workbench ║
╠══════════════════════════════════════════════════════════════════════════╣
║ 🔬 Open-BCI EEG → Text + Robotic Implementation [ACTIVE RESEARCH] ║
║ New architecture strategies · Commercialization development ║
║ 🧠 Cyborg AGI — Local AGI OS · Gemma 4 Hackathon · V1.0 Stable ║
║ MedGemma 4B · Adaptive Tutor · World Monitor · Mirofish Graph ║
║ 🌊 Flood-Guard AI — Hydrodynamic Sim · Satellite Validation ✅ Shipped ║
║ 🎭 Comic Companion — Geolocation Comedy Platform ✅ Shipped ║
║ 🍔 VIT-AP Eats — Campus Food Delivery PWA ✅ Live on Vercel ║
║ 📈 NewsAlphaAI — LangGraph Distributed Stock Prediction ✅ Shipped ║
║ 🧭 MeetMind — AI Teams Co-Pilot ✅ Shipped (Microsoft Hackathon) ║
║ 🛡️ CyberX — AgentOps Security Mesh ✅ Shipped (Security Hack) ║
║ 🎬 Editor Cyclops — Video Editing & Content Strategy ✅ Ongoing ║
╚══════════════════════════════════════════════════════════════════════════╝
