Crafting intelligent systems that think, reason, and learn.
☕ Turning caffeine into code.
> whoami --verbose
- 🎓 B.Tech CSE (Core) @ Global Academy of Technology, Bangalore — CGPA 8.62/10, Class of 2027
- 🧪 Turn research papers into working products — Explainable AI, RAG pipelines, and anomaly detection that actually ship
- 🧵 Juggling 8 concurrent "Lead Developer" projects and somehow all of them are still "Ongoing" (it's a feature, not a bug)
- 🎯 Two things obsess me: making AI explainable and making software fast — sub-150ms latency isn't a nice-to-have, it's a personality trait
- 🎤 Marketing Coordinator, IT Virtuoso Club — ran branding for HACK-A-LEAGUE and emceed 5+ tech events (turns out debugging code and hosting a stage aren't that different)
- 🧑💻 Ex-intern @ FutureInterns and Infosys; currently leveling up Generative AI chops via Google Cloud Skills Boost
- 📫 anuvutipodder205@gmail.com — I reply faster than my APIs (~250ms, allegedly)
Languages & Core
AI / ML
Web, Backend & Cloud
My core focus is Generative AI and LLM-powered systems — the tools orbiting the hub above (LLMs, RAG, vector databases, PyTorch, Hugging Face, Vertex AI) are the ones I reach for most, across Aven, InfraPredict, TalentIQ, and beyond.
Lead Developer building real-time Explainable AI (XAI) audit tooling.
- Established a scalable React 19 + TypeScript component architecture for live model-bias audits
- Built low-latency visual analytics (Recharts, Vega-Lite) mapping Statistical Parity & Disparate Impact without blocking the UI thread
- Shipped intersectional bias heatmaps and automated PDF export via html2canvas + jsPDF
Lead Developer on an exposure-therapy simulator combining conversational AI with real-time biometrics.
- Built an async NLP inference engine on fine-tuned RoBERTa (PyTorch/Hugging Face) for multi-label cognitive-distortion extraction
- Optimized background API brokers to keep round-trip latency under 1.2s for natural conversational pacing
- Integrated crisis guardrails, biometric tracking, and SUDS (Subjective Units of Distress) monitoring loops
Lead Developer on a cloud-native predictive maintenance pipeline for IoT infrastructure.
- Built FastAPI handlers ingesting multi-modal IoT telemetry under 50ms processing bounds
- Engineered unsupervised anomaly detection (Vertex AI, Isolation Forests) to shift monitoring from reactive to predictive
- Built an interactive GIS dashboard (Next.js + WebSockets) rendering live geographic health metrics
Lead Developer on an automated signature-verification and fraud-inference platform.
- Delivered end-to-end fraud inference under 150ms on a FastAPI backend
- Built a dual-layer pipeline: OpenCV micro-tremor analysis + a 97.8%-accurate soft-voting ensemble (SVM, RF, Gradient Boosting, LR, KNN) on HOG descriptors
- Added few-shot adaptive learning via online prototypical centroid updates to handle handwriting drift
| Project | What it does | Stack |
|---|---|---|
| CoreConnect | Multi-tenant HR/CRM workflow engine with sub-50ms indexed queries and enterprise-grade security (Helmet.js, rate-limiting, XSS filters) | Next.js · Node.js · Express · Prisma |
| SpaceIQ | Decoupled full-stack analytics dashboard with flexible MongoDB schemas and optimized REST endpoints | React 19 · Node.js · MongoDB · Tailwind |
| TalentIQ | Automated candidate evaluation engine cross-analyzing portfolios, resumes, and live GitHub APIs via OpenAI few-shot prompting | Python · OpenAI API · Hugging Face |
| AI Life Assistant | Face-recognition + RAG platform chaining OpenCV detection with vector-mapped knowledge retrieval (~5ms lookups) | Python · OpenAI · FastAPI · OpenCV |
| Ghost Dodger Game | 60 FPS offline-first PWA with a custom vanilla-JS game loop, collision detection, and Web Audio synthesis | HTML5 Canvas · Vanilla JS · PWA |
