public class PranavSagar {
String company = "Glance, InMobi Group — Bengaluru, India";
String role = "SDE 1 · Backend Developer (Sept 2023 – Present)";
String learning = "M.Tech AI & ML @ BITS Pilani (Work Integrated)";
String[] impact = {
"800K+ QPS · <30ms p95 latency · 100M+ users",
"35–60% compute & API cost reductions",
"Zero-downtime migrations via dual-write architecture",
"Reusable observability framework across 5+ services",
"90% reduction in deployment time via GKE automation",
};
String[] passions = {
"Distributed Systems",
"Event-Driven Architecture",
"Performance Engineering",
"Observability & Reliability",
"MLOps & Production AI",
};
}Languages · Backend · Messaging
Databases · Caching
Data Processing
Infrastructure · DevOps · Observability
🎯 Content Intelligence Pipeline — End-to-End MLOps
🌐 Live Demo · 🤖 API Docs · 🧪 MLflow · 📊 Grafana
Real-time news classifier serving fine-tuned DistilBERT at 94.64% accuracy with p95 latency of 23 ms on CPU. Production MLOps stack — streaming pipeline with at-least-once delivery, distributed caching with deduplication, drift monitoring, live dashboards, and automated CI/CD — all running on free-tier managed services with $0 recurring cost.
- 🧠 Model: DistilBERT fine-tuned on AG News (120K articles) · 94.64% test acc · 94.65% F1 macro · MLflow tracked
- ⚡ Serving: FastAPI on HuggingFace Spaces · async lifespan · full 4-class probability breakdown · p50 19 ms, p95 23 ms
- 🔄 Streaming: Redpanda Kafka → Redis cache (SHA-256 dedup) → SQLite · at-least-once delivery · 230 ms pipeline lag
- 📈 Observability: Prometheus client → Grafana Alloy (pull + remote_write) → Grafana Cloud · 6-panel live dashboard
- 🔍 Drift: weekly Evidently statistical tests (Jensen-Shannon + Wasserstein) → HTML report + MLflow on DagsHub
- 🤖 CI/CD: ruff lint + import smoke on every push · weekly drift cron · path-filtered auto-deploy to HF Spaces
- 📚 Docs: 12 ADRs · 15-problem build log · HLD + LLD with Mermaid sequence diagrams · layman newsroom analogy
|
NLP pipeline achieving 87% accuracy using Naive Bayes, SVM, and Random Forest classifiers. Full preprocessing with tokenization, stemming, and TF-IDF. Deployed as a real-time Flask web app. |
Scholarship disbursement platform eliminating 150+ km of rural travel per applicant. Built for Smart India Hackathon 2022 — National Runner-Up out of 1M+ participants. |
| 🏆 Avengers Award — Glance (InMobi) | Top engineering recognition for high-impact system design |
| 🥈 Silver Medalist — B.Tech CS, RTU | CGPA 9.10 / 10.00 |
| 🥈 Smart India Hackathon 2022 | National Runner-Up (1M+ participants) |
| 📜 Microsoft Certified | Azure AI & Data Fundamentals |
| 💻 LeetCode | Rating 1577 · Top 25% globally |
| ✍️ Published on Medium | Monte Carlo Simulation — most-read post |

