ABV-IIITM Gwalior · Information Technology · Class of '26
Software engineer focused on scalable backend systems, distributed infrastructure, and cloud-native observability — not just features.
My work spans production Kubernetes deployments, merged CNCF contributions, and full-stack GenAI pipelines.
🌐 CNCF Contributor Jaeger · OpenTelemetry · Kubernetes (PRs merged into K8s v1.36)
🏢 SDE Intern AlgoUniversity (YC S21) · Kubernetes · Redis · Docker · AWS
🎓 B.Tech Thesis UAV obstacle-aware path planning via Deep RL (D3QN / PPO)
🏆 McKinsey Fellow Forward Leadership Program
⚡ CP LeetCode Knight · 1200+ solved · 88% acceptance · 365-day streak
Code I've written runs in production across thousands of cloud-native deployments globally.
| Project | PR | What It Does |
|---|---|---|
| Kubernetes | #137398 |
Goroutine hot-loop fix merged into K8s v1.36 — client-go StartEventWatcher |
| Kubernetes | #137999 |
Fixed kubelet cgroup resize nil-pointer panic |
| OpenTelemetry | #46506 |
AST constant folding for pure OTTL functions — parse-time compiler optimisation, eliminates redundant per-datapoint evaluation in collector hot path |
| Jaeger | multiple | MaxTraceSize OOM safeguard · API v3 limit/attribute filtering · OTTL delete-matching-values · full unit test coverage |
GSoC 2026 · Proposed across 3 CNCF orgs: Jaeger AI Skills Framework · Kubeflow OTel Integration · Meshery AI Adapter
|
|
|
Production-grade RAG system over the GitLab Handbook — dual-path retrieval, SSE streaming, sub-200 ms latency
| Layer | Detail |
|---|---|
| Retrieval | Dual-path FAISS + BM25 with Reciprocal Rank Fusion (k=60) |
| Generation | Llama 3.3 70B via Groq · SSE streaming · 0–100 confidence scoring |
| Chunking | Heading-aware (350–500 tokens) · 10-rule generation protocol |
| Latency | Sub-200 ms retrieval · zero external API calls in search path |
| Infra | Single Docker multi-stage image (Node 20 + Python 3.9) · admin analytics · live re-ingestion |
Concurrent endpoint health-checking engine — retry logic, timeout propagation, bounded-latency execution
- Native Go concurrency for parallel monitoring with zero external dependencies
- Retry + timeout propagation ensuring bounded-latency under load
Full-stack competitive programming platform — Kubernetes-native, Redis-streamed, Docker-sandboxed
- cgroups + SECCOMP sandbox · Redis Streams decoupled eval pipeline
- Live contests · leaderboards · submission analytics · rating system
📌 LeetCode Knight · 1200+ solved · 88% acceptance · 365-day streak
📌 Flipkart GRiD Round 3 cleared
📌 AlgoU Camp Top 0.5% nationally
UAV-Assisted Obstacle-Aware Data Collection for Resource-Constrained IoT Networks
Supervisor: Dr. Ankur Jaiswal · ABV-IIITM Gwalior
- 800×600 m IoT deployment — 20 nodes, 3 cuboid obstacles (heights 128–148 m > UAV altitude 100 m)
- Greedy dominating-set selection → 65% node compression (20 → 7 stops, 100% coverage)
- Visibility-graph routing → collision-free paths (~2566 m, 9.2% battery)
- D3QN agent for adaptive path optimisation — state encodes UAV pose, residual energy, obstacle proximity, RP visit vector
- PPO as continuous-action extension
Published on Medium · System Design · Advanced DSA · Distributed Systems · SOLID · Caching · Graph Algorithms
"If it scales — it succeeds. Build with impact, not just intent."