▸ NLV-26-CRM Twenty requests, one row. Replies and bookings into a CRM exactly once. A Nordic B2B SaaS · with Tim N. 08/2026 → 09/2026 8 claims n8n workflows that move LinkedIn, email and Calendly events into the CRM without losing or doubling any of them, and a CRM switch with no downtime.
Problem Replies were missed, sent to Slack twice, or never reached the CRM. Build An atomic claim per event, CRM written before Slack, and a 24-hour shadow run. Result 0 errors in the shadow run and 2,075 duplicate companies merged.
Concurrent requests to rows written 20 → 1
Errors in the 24-hour shadow run 0
Duplicate companies merged 2,075 Open the record →
▸ NLV-26-AGT An agent pipeline that only says what its sources support Van Data Team · with Tran Tien Van 08/2026 4 claims Short tech-news videos made by a producer agent and its role agents, with every spoken sentence tied to a source and a person approving the hook, the script, the scenes and the final cut.
Problem Generated news videos drifted from their sources and nobody could say where a sentence came from. Build A producer agent over role agents, a claim registry, word-timing and clarity gates, and human approval at four stages. Result Runs stop instead of guessing, and every sentence on screen traces back to a source.
Role agents under one producer agent 8
Stages that wait for a person 4
Test cases in the files I added 302 Open the record →
▸ NLV-26-RIB Twenty account briefs that refresh themselves A US industrial supplier · with Tim N. 08/2026 → 3 claims · 1 public Sourced briefs on a sales team's target accounts and a Flash Report of what changed, rebuilt every 14 days. The public demo uses fictional names.
Problem Sales needed prioritized, sourced account research that stayed current. Build Deterministic Python builders, a relationship graph, scheduled refresh and a login. Result 20 briefs and a Flash Report, refreshed every 14 days.
Account briefs kept current 20
Test cases in the briefs repository 147 Open the record → Demo ↗
▸ NLV-26-KOL If Gmail times out mid-send, it checks Sent before retrying An influencer-marketing team in Vietnam 10/2026 1 claim A local-first outreach tool for recruiting creators. A language model drafts, a person approves, Gmail sends each email exactly once.
Problem One operator had to contact 100 to 150 creators a day without double sends. Build A SQLite job queue with leases, fixed Message-IDs and a check of the Sent folder. Result 23 of 23 acceptance cases passed with evidence.
Test cases in the KOL outreach tool 473
▸ NLV-26-TRK Renewal orders that reach Analytics, once each A subscription brand on Shopify, US · with Tim N. 07/2026 → 09/2026 3 claims Purchase tracking for a Shopify subscription brand. One Analytics home, ads that bid on real purchases, renewals and refunds sent from the server, and a daily check of every order.
Problem Sales went to two Analytics properties, ads counted cart steps as sales, and renewal orders did not reliably reach Analytics. Build One property, ads bidding on the purchase event only, renewals sent server-side with a key per order, refunds sent as refunds. Result 122 of 122 renewal orders found in Analytics on 7 and 8 September; none counted twice.
Renewal orders found in Analytics, 7 and 8 September 122
Renewal orders received server-side on 9 September, none counted twice 43 Open the record →
▸ NLV-26-GRO Leads that stopped disappearing A salon-suite studio in New York · with Tim N. 07/2026 → 2 claims Ad leads, CRM, SMS and cost-per-lead reporting for a salon-suite studio in New York.
Problem Leads from Meta ads and the website landed in tools nobody watched, and the SMS reply broke. Build Save first, filter only at the SMS step, send each form to four systems separately. Result Cost per lead tracked at $21.74 with spikes caught.
Cost per lead $21.74
Make operations saved per month 2,880
▸ NLV-26-OUT A cold-email program run like a system A US fractional-CFO firm · with Tim N. 07/2026 → 2 claims Daily health checks, inbox rotation, reply triage and positive replies routed to the CRM for a US finance firm.
Problem Deliverability and replies depended on someone checking inboxes by hand. Build Command-line tools on the sending platform's API, Slack reports and Zoho Flow routing. Result 114 inboxes checked daily and 234 replies classified.
Sending inboxes monitored daily 114
Test cases in the outbound tools 108
▸ NLV-26-SOC 3,370 posts in 34 minutes, and a checkpoint that looked into the future An AI social-scheduling SaaS · with Tim N. 07/2026 → 10/2026 2 claims A listening pipeline over 15 community sources that scores and de-duplicates posts and feeds a competitor study.
Problem The team needed to see relevant community threads every day without missing new ones. Build Scheduled scans with checkpoints, a SQLite ledger for duplicates and scoring. Result Duplicates down from 86% to 20%.
New posts collected in one run 3,370
Duplicate share in scans 86% → 20%
▸ NLV-26-ONT Ontevo, getting a paid AI product ready to charge for Ontevo (Lighthouse Insights) · with Tran Tien Van 04/2026 → 2 claims · 1 public Pricing, checkout and Stripe entitlements, tenant-safe customer simulations and admin tools for a US AI customer-insight SaaS.
Problem Paid plans, tenant safety and admin tools had to land in a large shared codebase. Build Next.js 15 screens and Django 5.2 services, each change shipped with tests. Result 637 commits and 293 test files added since 04/2026.
My commits on Ontevo 637
Test files I added to Ontevo 293 Open the record → Live ↗
▸ NLV-26-ATR Paid-ads attribution for WhatsApp leads A tutoring centre in Singapore · with Tim N. 08/2026 → Leads that arrived through WhatsApp ads had been counted as organic. An Apps Script engine matches them to ad sets and reports purchases back to Meta.
Problem Ad spend on WhatsApp leads looked like it brought nothing. Build Header-based column matching, a lock, hourly triggers and hashed Conversions API events. Result Spend and enrolments reconciled in one dashboard.
▸ NLV-26-XCL x-cli, a command-line client for X ↗ Own project 08/2026 → 09/2026 1 claim · 1 public Reads feeds and posts through the normal web interface; every write is a preview first, executed by ID.
Problem Agents needed to work with X without a paid API and without posting by surprise. Build A preview-then-execute model with immutable previews. Result 110 test cases, MIT licence.
x-cli test cases 110 GitHub ↗
▸ NLV-26-LCL lead-cli, search large lead datasets from the terminal ↗ Own project 08/2026 1 claim · 1 public Streams country files of 75 to 80 MB, filters, and writes NDJSON, CSV or TSV atomically. Ships as a single binary.
Problem Lead datasets were too big to open in a spreadsheet. Build A streaming parser, a blacklist check and cross-OS CI. Result 93 test cases on Ubuntu, macOS and Windows.
lead-cli test cases 93 GitHub ↗
▸ NLV-26-NEW InfoSaver, a Telegram bot that files links with a summary ↗ Own project 04/2026 Send a link to the bot; a language model summarizes it and the row lands in a Google Sheet by topic.
Problem Saved links were scattered across apps. Build Telegram bot, model summaries, Google Sheets as the store. Result 48 commits, in daily use. GitHub ↗
▸ NLV-26-UNI Multilingual lead routing, rules first Own project 10/2026 Leads from several channels are de-duplicated by phone, routed by language with rules for 7 languages, and an LLM is asked only when the rules are unsure.
Problem A sales team needed leads in seven languages routed to the right desk. Build Rules first, a language model as a fallback, first and last touch kept per customer. Result If the model fails, the lead is still routed and flagged.