Third-year Mechanical Engineering student at Jadavpur University who taught himself backend and systems engineering — WAL, LSM trees, gRPC, compaction — because that's what it took to build ZENITH from scratch. Now moving deliberately into AI infrastructure: agent frameworks, retrieval systems, and Kubernetes-native observability, rather than generic AI-app work. Core member of the JU ACM Student Chapter (AI & Blockchain division). Builds in public.
Open to: remote AI-infrastructure engineering roles · backend / distributed-systems internships
| Area | Evidence |
|---|---|
| Semantic search & retrieval | ZENITH — ONNX embeddings fused with BM25 via weighted RRF, Recall@10 0.812 → 0.906 |
| Agentic frameworks | Building Argus on the Mastra framework (agent orchestration + RAG pipeline) |
| RAG systems | Argus (in progress) — incident root-cause agent mesh, Mastra + pgvector |
| GSoC track | Targeting C2SI's b0bot for GSoC 2027 |
⚡ ZENITH — embeddable hybrid search engine, written from scratch in Go
Custom LSM storage engine (WAL → MemTable → SSTable → leveled compaction), inverted posting lists, BM25 lexical ranking fused with ONNX-based semantic retrieval via weighted Reciprocal Rank Fusion, served over gRPC with Prometheus instrumentation and a Cobra CLI.
| Stack | Scale | Performance | Repository |
|---|---|---|---|
| Go, gRPC, ONNX, Prometheus | ~120 commits · 29 stars · MS MARCO 100k passages | Recall@10 0.812→0.906 · BM25 latency 284ms→177ms | ZENITH |
🔍 Argus — incident root-cause agent mesh (in progress)
Standalone project (not built on ZENITH). Committed to a Mastra/TypeScript end-to-end stack with plain vector RAG via Mastra's built-in pipeline and pgvector as the vector store.
| Stack | Status |
|---|---|
| Mastra, TypeScript, pgvector | Architecture decided — build in progress |
| 📦 Repo | 🔧 What I did |
| internetarchive/Zeno | Race condition fix · PR #578 merged |
| supabase-community/supabase-go | SDK improvement · PR #51 merged |
| kyverno/kyverno (active) | Kubernetes policy engine (CNCF) |
| Recognition | Details |
|---|---|
| IEEE JUSB — DoubleSlash Hackathon | Finalist, 750+ participant field |
current_focus:
learning:
- Agentic AI architectures via Mastra
- Kubernetes-native observability patterns
building:
- Argus — incident root-cause agent mesh (Mastra + pgvector)
- Kubernetes SRE copilot template (Mastra ecosystem)
exploring:
- GSoC 2027 with C2SI (b0bot)
open_to:
- Remote AI infrastructure engineering roles
- Backend / distributed systems internshipsNow → Q3 2026 Mastra Kubernetes SRE copilot — merge + Argus build-out
Q3 2026 CKAD certification
Q3 2026 CKA certification
Q4 2026 LFX Mentorship Term 3 application
2027 GSoC 2027 — targeting C2SI (b0bot)
2027 Remote AI-infrastructure engineering role
Implementation notes on storage engines, search algorithms, and OSS contribution patterns — one post a month, no tutorials. Follow along on X (@shramanb113).
