dheeraj gajulaRésumé

Open to network & infra roles

I build networks that automate themselves and agents that debug them.

Dheeraj Gajula, network engineer. Production Engineer intern on Meta’s network team, Software Engineer at Versa Networks before that, and finishing an M.S. in Network Engineering at CU Boulder.

Get in touchHow I work
dc-fabric-01Healthy

all links up · 8/8 ECMP paths healthy

Illustration of the incident triage I built at Meta: agents read telemetry over MCP, find the cause, and act.
less on-call triage timeMeta
50%
requests per second servedVersa Networks
30k+
agents in one monitoring platformMeta
20+
billing turnaroundVersa Networks
7d → 1h

01How I work

The loop: design, automate, observe, diagnose.

Most of what I build sits somewhere on this loop. A network gets designed, turned into code, tested, deployed and measured, and when something breaks, agents help work out why. Pick a stage to see what that work looks like.

01 · CU Boulder network labs

Design the network

Redundant enterprise and ISP topologies: VLANs, STP, HSRP, multi-area OSPF, MPLS with RSVP-TE, IPsec over dual-stack IPv4/IPv6.

core-1.ios
interface Vlan10
 standby 10 ip 10.10.0.1
 standby 10 priority 110
 standby 10 preempt
!
router ospf 1
 network 10.10.0.0 0.0.0.255 area 0
 network 10.20.0.0 0.0.0.255 area 1

02Experience

Where I’ve shipped.

  1. May – Aug 2026

    Menlo Park, CA · Internship

    Meta, Production Engineer — Network

    Built the agents that investigate network incidents so on-call engineers don’t start from zero.

    • Built event-driven multi-agent orchestration on MCP tooling for WhatsApp root-cause analysis, cutting on-call triage time by 50%.
    • Designed a performance platform of 20+ agents with anomaly detection, monitoring WhatsApp, Instagram, Facebook and the CDN with automated reporting.
    • Engineered an MCP abstraction layer between network infrastructure and AI systems for deterministic tool-calling workflows.
    • MCP
    • Multi-agent systems
    • Python
    • Anomaly detection

  2. Jun 2024 – Aug 2025

    Bangalore, India · Full-time

    Versa Networks, Software Engineer I

    Backend services and data pipelines for an SD-WAN and security platform.

    • Developed high-concurrency REST APIs in Go and Cassandra serving 30,000+ requests per second, monitored with Prometheus and Grafana.
    • Analysed VirusTotal threat-feed data with Python and BigQuery and predicted threat patterns with ML models.
    • Containerised backend microservices with Docker and Kubernetes and deployed the pipelines to GCP.
    • Go
    • Cassandra
    • BigQuery
    • Docker
    • Kubernetes
    • GCP

  3. Feb – Jun 2024

    Bangalore, India · Internship

    Versa Networks, Software Engineer Intern

    Turned messy appliance logs into numbers the business could bill from.

    • Automated device-usage tracking, reducing billing turnaround from 7 days to under an hour.
    • Built anomaly detection over inconsistent logs, using state machines to track device state.
    • Modelled director-log telemetry with MongoDB and Python and surfaced health metrics in Prometheus.
    • Shipped Flask and FastAPI microservices with Docker Compose.
    • Python
    • MongoDB
    • FastAPI
    • Flask
    • Prometheus

03Projects

Labs I built, then broke on purpose.

2026 — ongoing · CSCI 5840 Advanced Network Automation

Enterprise network automation platform

A 16-node dual-stack enterprise network in Containerlab, run like production: a source of truth, CI-validated changes, live telemetry, drift recovery and tests that gate every deploy.

  • Tiered design on Arista cEOS and FRR: eBGP to an ISP, OSPF/OSPFv3 core, RIP at the edge with redistribution, VRRP gateways, stateful NAT with conntrackd session sync.
  • Changes start as YAML + Jinja2, are validated in an isolated three-vendor Containerlab matrix, promoted with SHA-256 manifests and deployed by Jenkins over Netmiko, with rollback.
  • Telemetry on four paths: SNMP and gNMI into Prometheus, traps and syslog into a MongoDB data lake, all in Grafana with email alerts.
  • Hourly drift detection and recovery, CPU alerting, daily credential rotation, a TACACS+ MFA pilot and NetBox IPAM.
  • 1,016 unit tests behind a custom coverage gate: 46/46 tools and 369/369 functions covered.
  • Containerlab
  • Arista cEOS
  • FRR
  • Jenkins
  • Netmiko
  • Jinja2
  • gNMI
  • SNMP
  • Prometheus
  • Grafana
  • MongoDB
  • NetBox
rcn-enterpriseGolden state

8 managed devices · running config matches golden

Replay of a recorded run (Sep 2026): S3’s OSPF cost changed in the source of truth, validated 129/129 across three vendor cases, deployed by Jenkins and read back from the device.

Fall 2026 · CSCI 5280 Software-Defined Networking · labs 6–8

AI in the SDN control loop

LLMs that read OpenFlow telemetry, turn plain English into validated policy and classify threats, with a human able to approve before anything changes.

  • AI Flow Analyst: a collector pulls flow and port stats from Ryu and Open vSwitch, an LLM analyses them as text and structured JSON, and an MCP server exposes it all to agents.
  • Natural-language controller: four typed MCP tools (block, allow, rate-limit, prioritize) produce schema-validated policy drafts and structured rejections.
  • Closed-loop response: detect port scans, UDP floods and unauthorized flows, classify them with an LLM, then drop or rate-limit in human-approval or automatic mode.
  • Ryu
  • OpenFlow 1.3
  • Mininet
  • Open vSwitch
  • MCP
  • LLMs
  • Python
  • MongoDB
sdn-closed-loopMonitoring

openflow 1.3 · flow + port stats polled via ryu REST

Lab 8: Ryu and Mininet with an LLM classifier and a response engine in human-approval mode. Switches lacked meter support, so rate limiting uses an OVS HTB queue.

Fall 2026 · CSCI 5280 Software-Defined Networking · labs 1–5

OpenFlow & SDN foundations

The groundwork under the AI work: OpenFlow down to the packet, controllers reaching switches across routed networks, and SDN apps people can operate.

  • Captured and decoded OpenFlow 1.3 handshakes, PACKET_IN/OUT and FLOW_MOD in Wireshark, including a version-mismatch failure.
  • A hybrid network: a Floodlight controller reaching Mininet switches across GNS3 Cisco routers.
  • REST-driven routing, firewall and QoS on Ryu, with VLAN flows verified from packet captures.
  • SDN security: caught a fake-switch PACKET_IN flood and auto-blocked only that connection with iptables, with host traffic unaffected.
  • A Flask operator app that drives static routing and a default-deny firewall through Floodlight’s REST API.
  • Midterm: Python automation from GNS3 OSPF to Ryu, with a live PACKET_IN dashboard.
  • OpenFlow
  • Floodlight
  • Ryu
  • Mininet
  • GNS3
  • Wireshark
  • Flask
openflow-1.3Connected

HELLO · FEATURES_REPLY · switch joined on OF 1.3

How a reactive OpenFlow switch learns a path: table miss, PACKET_IN, PACKET_OUT and FLOW_MOD, then forwarding without the controller. Decoded from Wireshark captures in lab 2.

Fall 2026 — ongoing · CSCI 5260 Datacenter Networks

Datacenter networks

How datacenters are wired and run, one hands-on lab a week: switching and spanning tree, MLAG, gateway redundancy, CLOS fabrics, VXLAN overlays, MPLS VPNs, load balancing, QoS, storage networks and VMware networking.

  • Switching on Arista: 802.1Q trunks, native-VLAN pitfalls, and STP timer tuning.
  • Port-channels and MLAG: LACP hashing tested with iperf3 (150 Mb/s across members shaped to 50 Mb/s each), MLAG peer-links and split-brain behaviour.
  • HSRP across four gateways and two VLANs, with load sharing between groups and interface tracking for failover.
  • MPLS L3 VPN on eight GNS3 routers: VPNv4 exchange with RD/RT design, a BGP-free core, and customer isolation verified in packet captures.
  • 802.1X with RADIUS-assigned VLANs and per-role router ACLs for a hotel LAN, with a guest portal.
  • CLOS fabric design and a study of Facebook’s fabric. Still ahead in the lab track: BGP load balancing, VMware networking and QoS with WRED.
  • Arista EOS
  • LACP
  • MLAG
  • STP
  • HSRP
  • MPLS
  • BGP VPNv4
  • 802.1X
  • GNS3
  • iperf3
asw6-asw8-poPo1 up · LACP

port-channel asw6 ↔ asw8 · four members, each shaped to 50 Mb/s

Replay of the CSCI 5260 port-channel lab on Arista switches. Each member was shaped to 50 Mb/s; the throughput figures are from our iperf3 runs, and the hosts are simplified.

Spring 2026 · CSCI 5180 Network Management & Automation

Network management system

A small NMS for a fleet of Cisco routers: it watches them, backs them up, and undoes changes that break the network.

  • Async Python SNMP poller behind a custom Prometheus exporter, with Grafana dashboards for CPU, memory and per-interface traffic and errors.
  • Flask + MongoDB backup service with a staged lifecycle: beta, golden running, golden and nightly configs.
  • Jenkins pipelines: a ping test after every change that promotes or rolls back on its own, and a nightly backup of every router.
  • Configuration and provisioning with Netmiko, NAPALM, Ansible, Jinja2 and NETCONF.
  • Python
  • SNMP
  • Prometheus
  • Grafana
  • Flask
  • MongoDB
  • Jenkins
  • Netmiko
  • NAPALM
  • Ansible
  • NETCONF
netmanGolden

configs move beta → golden running → golden · nightly snapshots at 00:00

Illustration of the CSCI 5180 config pipeline: a change that breaks reachability is rolled back to the last golden config. Three of the five routers are shown.

2025 — 2026 · CSCI 5170 IP Routing Protocols & Policies · enterprise and ISP labs

Routing: enterprise to Internet

How routes get chosen inside a network and between networks, configured by hand on Cisco routers and then failed on purpose.

  • RIP, EIGRP and single- and multi-area OSPFv2/v3 with redistribution, with failover timed from debug logs.
  • An ISP edge: route-maps per relationship (customer, peer, transit) and communities that decide which routes go where.
  • Customers attached statically and over eBGP, transit and peering with two upstream ASes, and BGP path selection and filtering.
  • VLANs, trunking, STP/RSTP and HSRP; MPLS with RSVP-TE tunnels for traffic engineering.
  • Dual-stack IPv4/IPv6 with IPsec tunnels and NAT-PT.
  • BGP
  • OSPF
  • RIP
  • EIGRP
  • Route-maps
  • MPLS
  • RSVP-TE
  • IPsec
  • IPv6
  • Cisco IOS
as1-edgeSessions up

route-maps {customer,peer,transit}-in and -out · communities record where a route came from

Illustration of CSCI 5170 Lab 3, built by a team of three: an ISP edge with customer, peer and transit policy written as route-maps and communities.

Fall 2025 · CSCI 5273 Network Systems

Network applications in C/C++

Protocol work from the socket up: servers, a caching proxy, a file-transfer protocol and a small distributed file system.

  • Multithreaded TCP/UDP servers and a custom UDP file-transfer protocol.
  • HTTP proxy with pipelining, persistent connections, MD5-keyed caching, prefetching and a URL blocklist.
  • Distributed file system with chunking, redundancy and MD5-based placement.
  • C / C++
  • POSIX threads
  • Sockets
  • HTTP
  • PostgreSQL
http-proxy.cListening

proxy listening · new thread per connection

The from-scratch HTTP proxy in C: a thread per client, MD5-keyed cache with expiry, and a URL blocklist. Illustrated sequence.

Before networking

I started in machine learning: research internships, a published paper, a national hackathon win. That’s why building agents came naturally later.

  • 2023–24Schizo AI Explainable ResNet over EEG signals for early schizophrenia detection. Published at ICADDI 2024.
  • 2022Tenali Legal-assistance engine for court backlogs. Won Smart India Hackathon 2022 on the DRDO track.
  • 2022–23Green Guru Crop recommendation and leaf-disease diagnosis for farmers.
  • 2024AppSynergies ML Research Intern. Ensemble models that cut fake profiles by 35%.
  • 2023–24IIT Bhilai Research Intern. Alternative credit scoring from transaction data.
  • 2023AutoSky Aerospace (IISc) Project Intern. +20% night-time detection with YOLOv8.
  • 2022Brane Enterprises Software Development Intern. Async ML validation backend with FastAPI and Redis.
  • 2022Amazon ML Summer School ML Apprentice. Competitive selection.

04Toolbox & training

What I reach for.

Protocols
  • TCP/IP
  • IPv4 / IPv6
  • BGP
  • OSPF
  • MPLS
  • RSVP-TE
  • STP / RSTP
  • DNS
  • DHCP
  • NAT
  • IPsec
  • SDN
Automation
  • Netmiko
  • NAPALM
  • Ansible
  • Jinja2
  • NETCONF
  • Jenkins
  • CI/CD
  • IaC
Languages
  • Python
  • Go
  • C / C++
  • Bash
  • SQL
Data
  • Cassandra
  • PostgreSQL
  • MongoDB
  • InfluxDB
  • BigQuery
  • Redis
Infrastructure
  • Linux
  • Docker
  • Kubernetes
  • GCP
  • AWS
  • NGINX
  • Prometheus
  • Grafana
Agentic AI
  • MCP
  • Multi-agent systems
  • Orchestration
  • Autonomous RCA
  • Tool calling

2025 – May 2027 · GPA 3.8

M.S. Network Engineering

University of Colorado Boulder

Enterprise Networks · Network Systems · SDN · Datacenter Networks · Advanced Network Automation · Machine Learning

2020 – 2024 · GPA 3.76

B.E. Computer Science & Engineering

Dayananda Sagar College of Engineering

Data Structures & Algorithms · Operating Systems · Computer Networks · Databases · Cloud Computing

  • WinnerSmart India Hackathon 2022 — DRDO track2022
  • Finalist & Best ContributorUNESCO India–Africa Hackathon2022
  • PaperExplainable AI in the Context of Schizophrenia — ICADDI2024

05Contact

Working on networks, infrastructure or agents for ops? Let’s talk.

[email protected]