I build tool-using AI agents and the backend they run on. Ask the console anything about my
work, or take the resume and go.
Ask about my work
Scripted replay. No model runs here; answers are pre-written from my resume and notes.
Every answer, written out
What agents has he built?
Three, at 10x Productivity.
Charlie is Archer's strategy agent, built on Claude. It uses tools, loads skills on demand,
gathers evidence in parallel, recalls earlier sessions and uses prompt
caching. The same tools sit behind an OAuth 2.1 MCP server, so the team can use them from
Claude or Cursor too.
The other two are workflow agents: a Meta lead-ads agent that sizes ad sets from the budget
and drafts on-brand ads, and a media-buyer agent that makes recommendations on ad
accounts.
How does he keep agents honest?
By keeping a person and a second check in the loop.
Nothing the lead-ads agent drafts goes live until a person approves the launch. The
media-buyer agent's recommendations are checked against computed figures and reviewed by a
critic model.
Before an ad launches, Creative X-Ray scores it against the brand's own guidelines and
lists what to fix.
What has he done with MCP?
Two MCP servers, both behind OAuth.
At 10x Productivity, an OAuth 2.1 MCP server exposes the same tools Charlie uses, so Claude
and Cursor can work with Archer's knowledge directly.
On the side, Letterport gives AI agents their own
email inboxes and ships a hosted MCP server with OAuth.
What's underneath Archer?
SourcesMeta Marketing API, Ad Library, AppMagic
IngestionBullMQ workers, Redis leader election
AnalysisScene-level video structure, 100+ signals
KnowledgeSQL analytics plus cached LLM insights
AgentsCharlie and the workflow agents
ClientsThe Archer app, Claude and Cursor over MCP
Ingestion pulls from the Meta Marketing API, Ad Library and AppMagic for 100+ competitor
brands, without a single client-specific pipeline. LLM cost is attributed by tenant and feature
across 60+ call sites.
It all runs on 7 ECS Fargate services, provisioned with Terraform and deployed through OIDC
CI/CD. He designed it and owns it end to end.
Did any of it move the business?
In a brand pilot, strategists accepted 85% of the AI-generated creative briefs they
reviewed.
The multimodal analysis has run on 15k+ ads, breaking video into scenes and scoring
100+ signals.
At BIK.AI he halved the Elasticsearch bill at 300M events a month, cut deploy time by
40%, and was named Best Employee of the Quarter in April 2025.
At BLKBOX.AI he went from SDE-1 to SDE-3 in under 2.5 years.
What does he build on the side?
Wellread shows a business what ChatGPT and Gemini
tell its buyers. It asks the questions buyers ask, with web search on, finds which brands
they name and drafts pages. 30+ early users so far.
Letterport is email for AI agents: inbound mail as
clean JSON with header-based threading, a per-recipient delivery trail, signed webhooks and
a hosted MCP server. Built with NestJS, AWS SES, SQS and CDK.
RapidMock lets frontend teams mock API responses
without waiting on the backend.
Where has he worked?
Oct 2025 – now
10x Productivity, Senior Software
Development Engineer, founding team
Jan – Oct 2025
BIK.AI, SDE-2
Aug 2022 – Dec 2024
BLKBOX.AI, SDE-1 to SDE-3
Jan – Jul 2022
HP R&D, software engineering intern
B.Tech in Computer Science from Pondicherry Engineering College, 2022. The resume has the details.
What's his stack?
TypeScript, Python and SQL. Node.js and NestJS with BullMQ on the backend; PostgreSQL,
Redis and Elasticsearch for data; AWS, Terraform and Docker for infrastructure.
On the AI side: tool calling, MCP, LangGraph, RAG and embeddings, LLM evals, multimodal
pipelines and structured outputs, across the Claude, Gemini and OpenAI APIs.
What's he like off the clock?
Usually on a bike, hunting down a good pizza, or halfway through a film.