5 years managing enterprise platforms under strict SLAs. 18 months self-funded R&D. 4 shipped systems. 17 open-source repos. One continuous arc from L2 support to multi-agent infrastructure.
In support, I owned L2 incident response for SAP modules and ServiceNow workflows. I learned what production discipline actually looks like under real SLAs, how non-engineers describe broken software, and where automation returns value.
In January 2024, I pivoted to writing systems. I built a multi-agent SDLC pipeline, a wellness ecosystem with Next.js and Expo mobile, a solo-founder outreach CRM, and an indie publishing bookstore.
I'm looking for a Forward Deployed Engineer or Technical Solutions role at an API-first startup. I bring operational reliability, debugger discipline, and systems-building chops.
Read my sabbatical story →Two live commercial services and two private betas. Click any card to review stack details, architectural bottlenecks, and known gaps.
Full-stack wellness platform syncing Next.js 15 and Expo 54 React Native. Employs hardware-encrypted storage (AES-256-GCM), log telemetry sanitization, and CrewAI consultations.
A 6-agent software development pipeline that reads a feature spec, structures plans, writes implementation tests, reviews syntax, and emits a PR description.
Client-facing outreach and CRM platform. Integrates Firecrawl lead enrichment, Supabase Edge Functions for multilingual messaging, and Instagram comment-to-DM triggers.
Independent publishing brand. Distributes three printed children's titles via Amazon KDP with a custom Shopify-like storefront direct to readers.
Seventeen repos, one continuous arc: understand the fundamentals, run production-grade solutions to the hard problems, then wire it all together. Pick your entry point.
Tier 1 · Learn
The same task solved at six levels of autonomy: code, single call, workflow, agent, agentic team, swarm. Run each rung in under a minute. Feel the jump.
One agent, four removable organs: brain (LLM), hands (tools), memory, loop. Toggle one off and watch it break. The fastest answer to “what is an agent, really?”
Tier 2 · Solve
Stops agents from failing the same way twice. Hashes failure signatures, stores the resolution, and injects the known fix on the next occurrence. Extracted from 18 months of AgentKernel production incidents.
Eliminates blind-spot refactors. Before any agent edits a file, maps the full import graph and surfaces every downstream module at risk. Prevents cascading failures in autonomous coding pipelines.
Cuts repeat-problem resolution time to near zero. Stores successful solutions by context hash and retrieves them across sessions, injecting the proven approach before the agent re-invents it.
Zero downtime LLM routing across 9 providers. When one API degrades, the circuit opens and traffic fails over automatically in under 50ms. Tested across 13 failure scenarios including cold-start and simultaneous outage.
Tier 3 · Production
Cuts deep research from hours to minutes. Queries 4 search providers in parallel, scores sources for relevance, and produces a cited structured brief. Runs on any LLM via the 10-provider router.
Turns raw URLs and unstructured text into cited structured JSON in seconds. The intake step for automated content pipelines that process hundreds of documents per run.
Text-to-video in under 2 minutes. A brief becomes a narrated, frame-synced 9:16 MP4 without touching video editing software. Temporal Authority ensures audio and frames never drift apart.
Catches agent drift before it reaches users. Define behavioral rules once, then run automated checks that compare live agent output against the original constraints. Built after a production drift incident.
Connect any Claude or GPT agent to production tools via the Model Context Protocol. Exposes blackboard state, SCAR failure lookup, and LLM routing as MCP-standard tools any agent can call.
Ask questions about any codebase or document collection and get cited answers in seconds. Ingests, chunks, embeds into Qdrant, then retrieves with semantic search before Claude generates the answer.
Tier 4 · Integrate
Prevents the architecture mistakes that cost weeks to undo. Seven production-grounded patterns from DAG scheduling to prompt-injection defense, each traced back to the real failure that motivated it.
Five working agent workflows you can run offline in under 5 minutes. Each is a self-contained learning lab: swap the LLM, break a component, and watch it fail gracefully. Ollama-compatible, zero API key needed.
18 months of production incidents distilled into six modular backend engines. Every SOLVE repo in this portfolio was extracted from here. The infrastructure layer that keeps running when things go wrong.
Launch a billable multi-agent SaaS in a weekend. SSE progress streaming, DAG scheduler, and geo-routed Stripe/Razorpay billing wired in from day one. 22 tests passing. Extracted from a live commercial system.
The narrative spine and template that links all Solve Tier repositories into a single production workflow.
Owned global escalation queues for SAP modules and ServiceNow workflows. Automated triage log checks using JavaScript, compression troubleshooting, and maintained target satisfaction ratings.
Full operational path →Built Next.js 15 mobile integrations with secure storage enclaves, async LLM routing templates with circuit breakers, and custom topological scheduling engines for multi-agent networks.
Sabbatical execution details →I respond within one business day. Feel free to reach out for a code walkthrough or screenshare of any production beta systems.