Hokma Solutions / Case study
Support Coach: a pre-send reply checker
Commercial product. It is live on the Chrome Web Store, with a production backend.
My role. Founder, product architect and verification lead (implementation assisted by AI coding tools under strict test rules)
Stack: Chrome Extension MV3, TypeScript, Vite + CRXJS, Node.js + Express, Supabase PostgreSQL, OpenAI (gpt-4o-mini), Paddle Billing, PostHog (Server-Side)
Landing page (supportcoach.io/extension)Chrome Web Store listing60-second product demo
The problem
Live-chat support agents under queue pressure sometimes type blame-heavy, dismissive, or dead-end replies ('Like I said, you need to reconnect it yourself'). Built-in help desk AI tools only help when the agent clicks a button to ask, which a rushed, confident agent never does. Post-hoc QA tools only flag the damage days later.
How it was built
- Built a 3-layer coaching engine: (1) an instant browser-local rules engine with no network call, no AI cost, and offline support, (2) a paused-typing AI coaching card, and (3) a manual pre-send AI check with a one-click rewrite.
- Built a platform-adapter DOM layer for Zendesk Agent Workspace (contentEditable), Intercom Inbox (contentEditable), and Zoho SalesIQ (<textarea> with synthetic input/change event dispatch and edit-box focus guards).
- Fixed real-world DOM failures found through field diagnostics: replaced naive CSS-class modal checks, which broke on chats with file attachments, with geometric overlap and elementsFromPoint z-order visibility gates.
- Found why agents ignored early local hints using a written test set with no customer data: two absence-based detectors fired on plain factual replies, and clicking Accept replaced the agent's factual answer with a generic opener. Added co-occurrence gating and non-destructive sentence stitching.
- Enforced strict privacy boundaries: the manifest requests only storage and five help desk host patterns; only draft_text, triggered_rules, local_score and draft_hash leave the browser (customer messages are never read); no third-party SDKs in the extension.
Results I can show
- Support Coach is live on the Chrome Web Store.
- The rules engine, the AI layer, the backend and the extension each have their own automated tests.
Limits
- No consulting client references yet.
- No customer satisfaction, escalation-rate or resolution-time numbers are claimed, because no controlled before-and-after has been run.
- Reads only the agent's draft (never customer messages), so rewrites cannot infer unstated customer questions.
Try the demos
Synthetic data. The local rules engine is the real deterministic code. The AI layer here returns sample output. No model is called.
Layer 1 (local rules) runs the real deterministic code. The AI layer here returns sample output. No model is called.
Fields that leave the browser (apps/extension/src/lib/apiClient.ts)
{
"draft_text": "Like I said, you need to reconnect it yourself. It's not a bug on our end, and there is nothing I can do from here.",
"triggered_rules": [
"dismissive_harsh",
"blame_language",
"missing_ownership",
"cold_transactional"
],
"local_score": 72,
"draft_hash": "ead21245a262ef2d"
}
Layer 1: local rules (no AI)
Local score: 72/100
This could read as dismissive or abrupt to the customer.
Example: "I can help get this working for you right now. Let me walk you through the steps."
Rule families fired: dismissive_harsh, blame_language, missing_ownership, cold_transactional
Layer 3: pre-send check (sample output)
Overall risk: high
Identifies the reconnection requirement needed to resolve the issue.
- empathy
- 2/10
- ownership
- 2/10
- clarity
- 9/10
- actionability
- 9/10
- professionalism
- 2/10
- de escalation
- 2/10
- conciseness
- 9/10
Suggested rewrite
I can help you get this reconnected. This happens when the local connection token resets rather than a platform outage, and I can walk you through reconnecting it right now so everything syncs cleanly.
The deterministic parser, screen-share gap exemption (5 minutes or more after a session link), neutral scoring of abandoned chats, and the topic schema from supportcoach-ai, run on synthetic transcripts.
Attention: high
- Orphan Line & Sender Normalization (Task 8e/8h): Multi-line messages preserved under active speaker.
- Response-Time Threshold Rule: Gaps under 2 min ignored; only gaps >4 min without screen-share URLs are cited (max 2-3 timestamps).
Coaching note for the manager
Quick note on this chat: you got the customer to the right place (Settings > Billing). For next time, lead directly with what you CAN do ("I can reissue that invoice with your new address right now") rather than opening with "That is not possible", and skip phrases like "As I said earlier" after a 4.1-minute gap.
Product confusion topics
- Billing: update billing address on past invoice (confusion, 94%)
"Hi, I'm trying to download last month's invoice with our new company address, but the PDF keeps showing the old billing address." - Billing: update billing address on past invoice (confusion, 94%)
"Wait, isn't there a way to regenerate it? And why didn't the address update?"
A simulation of the two-tap approval card and the multi-provider pace router, on synthetic tasks. The agent's prediction stays hidden until you decide.
Pace router
- Anthropic (Claude Sonnet): 54% used this week, resets in 52h (behind pace (-14 pts))
- Google (Gemini Flash / Gem): 38% used this week, resets in 52h (behind pace (-30 pts))
- OpenAI (gpt-4o-mini API): 79% used this week, resets in 120h (ahead of pace (+11 pts))
sc-sender · effect: external (medium)
Send today's outbound batch: 4 verified prospects
The agent's prediction is committed and hidden until you decide.
Fifteen minutes, and you will know if it's a fit.
Tell me what your team is dealing with. I will tell you whether I can help and what it would cost.
Prefer to write?
Send a short brief