chamath/
available·responds in ~4h
// senior engineer · automation · agents

I build AI systems
that don't break in
production.

Ten years shipping Angular, React, and Node at scale — now applied to n8n, LangChain, and multi-agent workflows. The kind of automation you can actually trust with real money and real users.

13+ years
shipping production code
// from trainee to lead, no gaps
5 companies
startup to enterprise
// codegen, medisoft, conxillium, apium
~40 workflows
deployed in last 12 months
// n8n, make, custom agents
100%
built and maintained by me
// no offshore team, no handoff
// what I actually do

Two services. Both engineered.

scope · spec · ship
[01]/automation

n8n · Make · Zapier + LLMs

Lead capture, document processing, AI-assisted ops, internal copilots — engineered like backend services, not duct-taped together.

  • CRM enrichment pipelines
  • RAG over your docs
  • Multi-step approval flows
  • Slack / email triage agents
  • Scheduled report generation
typical engagement: 2–6 weeks · from $4,500
[02]/agents

Multi-agent systems, autonomous workflows

Sales agents that close, support agents that don't hallucinate, ops agents that actually do the work — with guardrails and observability.

  • Tool-using agents with LangGraph / CrewAI
  • Voice + chat interfaces
  • Human-in-the-loop guardrails
  • Observability and eval pipelines
  • Custom MCP servers
typical engagement: 4–12 weeks · from $12,000
// selected work

Recent builds.

6 of 40+ · last 24 months

ThirdPay

thirdpay·web3 · business wallet·2024·angular · ethers · stripe

// problem

B2B teams paying suppliers in stablecoin had to bounce between Coinbase, internal sheets, and ERP. No audit trail, no approvals, no sleep.

// solution

Built the production payments console in Angular — multi-sig approvals, on-chain settlement, real-time reconciliation against the company's books. Live with paying customers.

−87%
manual reconciliation time
$2.4M
settled in first quarter
4-eyes
every transaction
angulartypescriptethers.jsstripepostgres

Lendermate

lendermate·fintech · credit ops·2023·react · node · aws

// problem

Underwriters were pasting bank statements into spreadsheets, then waiting two days for risk to sign off. Pipeline was bleeding deals.

// solution

Replaced the manual flow with a document ingestion pipeline + risk dashboard. OCR, categorization, scoring, and approval routing — all auditable.

48h → 2h
time to decision
12k/mo
documents processed
+34%
qualified application throughput
reactnodelambdatextractpostgres

B2B SaaS — SDR Triage Agent

[confidential]·saas · revenue ops·2025·langgraph · openai · hubspot

// problem

SDR team was spending 60% of their day qualifying inbound leads — most of which were unfit. Reps burned out, real opportunities went cold.

// solution

Shipped a LangGraph agent that researches each lead (firmographics, intent, fit signals), drafts the first reply, and only escalates to a human when scoring crosses threshold. Eval suite catches regressions before deploy.

−60%
SDR triage hours
3.4×
qualified-meeting rate
$0 → $40k
MRR pipeline / quarter
langgraphopenaihubspotclearbitnode

Legal — Contract Pipeline

[confidential]·legal · operations·2025·n8n · claude · pinecone

// problem

Mid-size firm was reviewing 12k contracts/month manually — clause extraction, risk flagging, redline drafting. Three paralegals couldn't keep up.

// solution

n8n pipeline pulls contracts from intake, Claude extracts clauses + flags non-standard language, Pinecone surfaces near-duplicates from precedent. Paralegals review surfaces, not raw PDFs.

12k/mo
contracts processed
−74%
review time per contract
100%
audit trail per clause
n8nclaudepineconenodes3

Internal Copilot — Apium

apium innovations·internal tools·2024·n8n · openai · slack

// problem

Engineering kept getting pinged about deploy status, incident timelines, and 'what changed when.' Knowledge was scattered across Slack, Linear, and the deploy logs.

// solution

A Slack-native copilot that answers ops questions by querying Linear + deploy logs + Slack history through a RAG index. It cites sources. People stopped @-ing engineers.

−42%
interrupt pings to eng
<3s
median answer latency
94%
answers with cited source
n8nopenaipineconelinearslack

MC Medisoft — Claims OS

mc medisoft·healthcare · claims·2021·angular · .net · sql

// problem

Claims processors were toggling between four legacy apps to settle a single claim. Error rate was 11% — and every error was a callback.

// solution

Rebuilt the operator console in Angular as a single workspace. Validation moved to the front-end, server enforced invariants, and a thin event log gave audit visibility for compliance.

11% → 1.6%
claim error rate
3.2×
claims processed per FTE
0
compliance findings YoY
angularrxjs.netsqlserver
// free workflows

Steal these. They work.

Battle-tested n8n and Make templates. Drop in, swap your keys, ship. No email required — but I'd love to know what you build.
gmail-to-notion-triage.json

Auto-classifies inbound Gmail, drafts a reply, and files the thread into the right Notion DB.

n8n · openai · 12 nodes
[] intermediate248
stripe-to-slack-digest.json

Daily revenue digest in Slack — MRR delta, top churners, new logos. Zero queries to write.

make · stripe · 8 nodes
[] starter612
rag-over-google-drive.json

Watches a Drive folder, chunks new docs into Pinecone, exposes a chat endpoint. Drop-in RAG.

n8n · pinecone · 21 nodes
[] advanced184
lead-enrichment-apollo-claude.json

Takes an email, returns firmographics + a written-by-Claude opener. HubSpot ready.

n8n · claude · 14 nodes
[] intermediate327
contract-clause-extractor.json

PDF in → structured JSON out, with non-standard clauses flagged. Plug it into your review queue.

make · claude · 9 nodes
[] intermediate142
github-pr-summary-to-linear.json

Every merged PR posts a human summary to the Linear issue it closed. Stops the 'what shipped?' Slack thread.

n8n · openai · 7 nodes
[] starter401
// the difference

Most “AI consultants” can't read their own code. I can.

// most consultants
glue zapier blocks until something works
// what I do
write custom nodes when no-code hits the wall
// most consultants
hand off and disappear
// what I do
instrument, monitor, and iterate post-launch
// most consultants
"AI is magic"
// what I do
cost-per-run analysis before you commit
// most consultants
prompt and pray
// what I do
eval pipelines, regression tests, guardrails
// most consultants
their stack, their lock-in
// what I do
your infra, your keys, your data
// 13 years of receipts

The log.

click to expand
2024 → presentlead engineerapium innovations
  • Leading AI automation practice — n8n, LangChain, custom agents for SaaS + fintech clients.
  • Shipped the internal ops copilot (Slack + RAG over Linear & deploys) cutting interrupt pings 42%.
  • Stack: typescript, node, langgraph, n8n, openai, anthropic, pinecone, aws.
2023 → 2024senior engineerconxillium group b.v.
  • Owned the front-end platform for the energy-trading dashboard — Angular + RxJS at scale.
  • Built the first internal LLM integrations (document summarization, knowledge retrieval).
2018 → 2023senior engineermc medisoft
  • Rebuilt the claims-processing console — claim error rate from 11% to 1.6%.
  • Mentored four juniors; introduced design-system + testing discipline.
2013 → 2018software engineercodegen international
  • Built airline-revenue products for tier-1 carriers. Angular, .NET, SQL Server.
  • First production code shipped: 2013. Still in flight.
2012 → 2013trainee engineercodegen international
  • Front-end intern. Where the typing-out-loud started.
// currently most fluent in:n8n·make·langchain·langgraph·openai·anthropic·pinecone·react·angular·typescript·node·aws·postgres·mongodb
// how we work together

Four steps. No surprises.

[01]
discovery
30-min call, no slides, just questions about what's actually broken.
[02]
scoping
Fixed-price proposal with deliverables, timeline, and one explicit 'what this isn't.'
[03]
build
Weekly demos, async updates, you see it work before you pay the final invoice.
[04]
handoff
Documentation, training session, 30 days of support included.
// let's build something

Tell me what's broken.

Tell me what's broken, what you've tried, and what 'done' looks like. I read every email — usually reply within 4 hours during Sri Lanka daytime (GMT+5:30).