Skip to content

Swapnil Surdi

I build production AI systems — RAG pipelines, agentic fleets, and the backend infrastructure that keeps them fast, cheap, and reliable.

[email protected] github linkedin

100K+

req/day @ 99.9%

30s → <1s

RAG retrieval

−30–50%

LLM cost (mcp-cache)

8min → 30s

MRI loads

02 — activity

Activity

github · swapnilsurdi

16 repos · 9 stars · 698 contributions/yr

julsmtwtfs2026-07-01: 0 contributions2026-07-02: 0 contributions2026-07-03: 0 contributions2026-07-04: 0 contributions2026-07-05: 0 contributions2026-07-06: 0 contributions2026-07-07: 0 contributions2026-07-08: 0 contributions2026-07-09: 0 contributions2026-07-10: 0 contributions2026-07-11: 0 contributions2026-07-12: 0 contributions2026-07-13: 0 contributions2026-07-14: 0 contributions2026-07-15: 0 contributions2026-07-16: 0 contributions2026-07-17: 0 contributions2026-07-18: 0 contributions2026-07-19: 0 contributions2026-07-20: 0 contributions2026-07-21: 0 contributions2026-07-22: 0 contributions2026-07-23: 0 contributions2026-07-24: 0 contributions2026-07-25: 0 contributions2026-07-26: 0 contributions2026-07-27: 0 contributions2026-07-28: 0 contributions2026-07-29: 0 contributions2026-07-30: 0 contributions2026-07-31: 0 contributionsaugsmtwtfs2026-08-01: 0 contributions2026-08-02: 0 contributions2026-08-03: 0 contributions2026-08-04: 0 contributions2026-08-05: 0 contributions2026-08-06: 0 contributions2026-08-07: 0 contributions2026-08-08: 0 contributions2026-08-09: 0 contributions2026-08-10: 0 contributions2026-08-11: 0 contributions2026-08-12: 0 contributions2026-08-13: 0 contributions2026-08-14: 0 contributions2026-08-15: 0 contributions2026-08-16: 0 contributions2026-08-17: 0 contributions2026-08-18: 0 contributions2026-08-19: 0 contributions2026-08-20: 0 contributions2026-08-21: 0 contributions2026-08-22: 0 contributions2026-08-23: 0 contributions2026-08-24: 0 contributions2026-08-25: 0 contributions2026-08-26: 0 contributions2026-08-27: 0 contributions2026-08-28: 0 contributions2026-08-29: 0 contributions2026-08-30: 0 contributions2026-08-31: 0 contributions
lessmore
claude code · this machinepeak 467m

5.3b tokens total · ~146m/day (30d avg)

julsmtwtfs2026-07-01: no activity2026-07-02: no activity2026-07-03: no activity2026-07-04: no activity2026-07-05: no activity2026-07-06: no activity2026-07-07: no activity2026-07-08: no activity2026-07-09: no activity2026-07-10: no activity2026-07-11: no activity2026-07-12: no activity2026-07-13: no activity2026-07-14: no activity2026-07-15: no activity2026-07-16: no activity2026-07-17: no activity2026-07-18: no activity2026-07-19: no activity2026-07-20: no activity2026-07-21: no activity2026-07-22: no activity2026-07-23: no activity2026-07-24: no activity2026-07-25: no activity2026-07-26: no activity2026-07-27: no activity2026-07-28: no activity2026-07-29: no activity2026-07-30: no activity2026-07-31: no activityaugsmtwtfs2026-08-01: no activity2026-08-02: no activity2026-08-03: no activity2026-08-04: no activity2026-08-05: no activity2026-08-06: no activity2026-08-07: no activity2026-08-08: no activity2026-08-09: no activity2026-08-10: no activity2026-08-11: no activity2026-08-12: no activity2026-08-13: no activity2026-08-14: no activity2026-08-15: no activity2026-08-16: no activity2026-08-17: no activity2026-08-18: no activity2026-08-19: no activity2026-08-20: no activity2026-08-21: no activity2026-08-22: no activity2026-08-23: no activity2026-08-24: no activity2026-08-25: no activity2026-08-26: no activity2026-08-27: no activity2026-08-28: no activity2026-08-29: no activity2026-08-30: no activity2026-08-31: no activity
lessmore

as of 10 jun 2026

03 — selected work

Selected work

All projects →

▣ live · 3 nodes · 22 containers

LaunchLab Fleet

Three recycled laptops, each operated by its own headless Claude Code agent: a private 22-container homelab that monitors, heals, and reports on itself.

  • 288 watchdog runs/day, zero tokens
  • 4.18s → 18ms status query
  • 22 containers

▣ npm · @hapus/mcp-cache · ★9

MCP-Cache: a transparent cache for any MCP server

A transparent proxy that caches oversized MCP tool responses and hands the model query tools — so any MCP server works past the 25K-token wall.

  • 25K → unlimited token wall
  • −30–50% LLM API cost
  • <200ms cached query

▣ production · HIPAA · 4 yrs

Agentic RAG in regulated healthcare

Production agentic RAG over docs, code, Confluence, and Jira for a HIPAA/ISO 13485 platform — compliance retrieval 30s → sub-second, verification 60% faster.

  • 30s → <1s compliance retrieval
  • 60% faster verification

04 — writing

Writing

All posts →

41× from one keyword

A FastAPI service on a fixed 1 vCPU went 1.68 to 69.6 RPS by adding async — before any hardware, workers, or DB tuning. A staged k6 study of throughput.

  • #python
  • #fastapi
  • #async
  • #performance
  • #benchmarking

Evals before vibes

Choosing an LLM by feel ships regressions you can't see. Picking models with an eval framework instead — latency, cost, accuracy, fit — from production.

  • #llm-evals
  • #rag
  • #llmops
  • #production-ai

Three laptops, one subscription

A fleet pattern for 24/7 AI agents: one agent per machine as gatekeeper, a star topology, a chat room as the bus, and a subscription instead of metered keys.

  • #ai-agents
  • #claude-code
  • #fleet
  • #self-hosted
  • #architecture

Looking for the full picture — roles, stack, and the numbers behind the work?

View resume →