Forward-Deployed AI Product & Solutions Operator

Raj Ranpariya

HVAC comfort advisor and project manager who builds AI systems.

Customer-facing technical sales for homes and commercial sites, project ownership from the signed proposal to the final inspection, and AI systems built by directing AI rather than hand-writing code: that is the combination I bring. I use AI to get more out of every part of life, professional and personal, and I verify what I build against running systems rather than trusting a report that says it works.

Agent orchestration mapCapability nodes — call handling, scheduling, work orders, permits, accounting, sales follow-up, reviews, marketing, reporting — connected by hand-offs around an operator hub. Red nodes are human approval gates and an external verification step.operatorcall handlingschedulingwork orderspermitsaccountingsales follow-upreviewsmarketingreportinghuman gatehuman gateexternal check
Agent orchestration mapCapability nodes — call handling, scheduling, work orders, permits, accounting, sales follow-up, reviews, marketing, reporting — connected by hand-offs around an operator hub. Red nodes are human approval gates and an external verification step.operatorcall handlingschedulingwork orderspermitsaccountingsales follow-upreviewsmarketingreportinghuman gatehuman gateexternal check
  • agent
  • human gate / external check
  • operator
Where the humans are is the design. The model handles judgment; every one-way door has a gate.
RR

01About

About

I sell and manage heating and cooling projects for a living, for homes and commercial sites, and in 2026 I started building software seriously: agents, orchestration, and the verification around them, first as personal tools and then as a full multi-agent system built as an independent project.

More

I build with AI as an engineering partner. I define the problem, the workflow, the acceptance criteria and the human gates; I direct the implementation, then verify it against the running system rather than trusting a report that says it worked.

Engineering background: MS in mechanical engineering, New York Institute of Technology, 2018; BE in aeronautical engineering, 2016.

02The combination

Customer understanding, technical sales, domain workflow, AI systems, orchestration, verification

I sell heating and cooling systems and I build AI systems, and each one makes me better at the other.

More

In sales I earn a customer's yes and then own the project until the inspection passes. At the desk I build agents and check them against running systems, not against reports that say they work. The sales work teaches me what a customer actually needs to hear; the engineering teaches me what a promise really costs. This site is a worked example of the same habit in every part of life: my own judgment plus artificial intelligence, applied to sales, projects, engineering, money, health and the paperwork of being a person.

03Skills

What I can actually do

Sales and customer-facing

Discovery with homeowners and business owners, consultative selling, financing conversations, objection handling, closing, post-sale ownership until the inspection passes.

Project management

Heating and cooling load analysis, system design and right-sizing, permits and inspections, installation coordination, technician ride-alongs, crews, inspectors and municipalities.

HVAC and building science

Test-and-balance, airflow, combustion safety, home-performance diagnostics, retrofit and new construction, light-commercial systems.

AI systems, built by direction

I do not hand-write code. I define the problem, the workflow, the acceptance criteria and the human gates; I direct AI to build it, then verify the result against the running system. Agent orchestration, voice-to-action, human-in-the-loop gates, evaluation harnesses, retrieval with provenance.

Hosting and infrastructure

Linux servers and Docker, databases and vector stores, local open-weight models, DNS, domains and edge hosting, firewalls and backups, webhooks and API integrations: chosen, set up and kept running.

04Projects

Built to learn, verified against the running system

Projects I built to learn and to demonstrate what I can do, mostly in the problem space I know best: field-service operations.

More

Every one was built through directed AI-assisted implementation: I defined the problem, the workflow, the gates and the acceptance criteria, directed the implementation, and verified the result against the running system. Descriptions say what runs today and what is only designed; nothing here is a product or a business.

Each card links to its proof: design notes, diagrams and the real verification harnesses on GitHub, with product code kept private.

Comparison and presentation tool for the sales conversation

A presentation tool for comparing heating and cooling systems on the three things a customer actually feels: comfort, efficiency and sound.

More

One self-contained HTML file for a tablet, with embedded photos, drawings and unit sound demos calibrated so the differences you hear are proportionally real, a comfort-and-performance ladder that builds as you add systems, plain-language explainers on right-sizing, humidity and warranties, and a one-tap PDF summary the customer can take away. Built by a salesperson for the sales conversation; it works offline and travels by AirDrop.

ProofDesign notes + diagram

Field app with a 3D crew packet

A single-purpose iPad app for the sales appointment, built for my own day job.

More

It replaces the paper comfort-survey booklet with a seven-step flow: the job, a 35-question comfort survey, an observation sheet for the existing equipment and ductwork, typed tape-measure dimensions, a live rotatable 3D model built from those numbers, a 16-shot photo set and an optional signature. It exports one self-contained crew packet the install team can review remotely, which removes the separate pre-install walkthrough. It runs fully offline with no backend, and it has two honesty rules I refuse to improve away: photos never generate the 3D, and a blank field stays visibly blank rather than being guessed.

ProofDesign notes + diagram

Backorder Buster

Backorder Buster, a small automation for my day job that attacks the most expensive surprise in a sales pipeline: a unit that is sold, scheduled and then turns out to be backordered.

More

It signs in to the distributor portal with a captured session, checks the availability of every model on the upcoming installs, classifies each one as in stock, low stock, backordered or restricted, and emails a short report before the install date is at risk, so a model can be swapped or a customer warned while there is still time. Built because I was the person who had to make that phone call.

ProofDesign notes + diagram

Voice to structured action

Voice-to-structured-action layer for a field-service operations system.

More

A seven-assistant squad: one call-taker greets, withholds the caller's name until confirmed, triages urgency from a per-trade knowledge pack, and hands off to six specialists. The voice platform's 30-second budget per tool call forced a split: anything on the conversational path runs in-process against the database; slower work falls to a workflow engine behind a 25-second abort. Caller identity comes from telephony metadata and overwrites whatever the model supplies; a booking exists only when a row is written. A weekly text-simulation harness runs three scripted scenarios across 18 trades; the latest run passed 18 of 18.

ProofDesign notes + diagram

Scheduling and work orders

Scheduling and work-order agents.

More

One workflow-engine agent books, reschedules and cancels appointments against existing bookings; the other turns a won job into a work order that a manager must review before the scheduling and billing handoffs fire. Every request passes authentication, a role check and per-tenant task toggles; installation, inspection, callback and emergency bookings are held for human approval. The voice channel offers slots filtered against existing appointments and has a review-first mode for busy periods. A daily automated check exercises 14 agents' write paths and asserts a row persisted; the last seven runs passed 14 of 14. Dispatch recommendations remain stubs.

ProofDesign notes + state diagram

Permit and inspection lifecycle

Permit-and-inspection lifecycle agent.

More

One webhook entry point routes six actions behind authentication and role checks; permit status is an explicit state machine that spans inspection scheduling and results, unknown actions return 400, and a locked design rule says nothing is ever auto-submitted. A headless-browser sidecar, memory-capped and loopback-only, with form filling, CAPTCHA detection and a WebSocket hand-off to a human, is built and health-checked, but no workflow calls it yet and only test screenshots exist. What runs today is record creation plus a deterministic permit-required decision, exercised by a daily automated check.

ProofDesign notes + diagram

Post-job follow-up: accounting, reviews, referrals

Post-job follow-up agents, the pair I am most pleased with because the timing is the product.

More

An accounting agent creates invoices, reports collections, and exposes four payment-exception types through a manager-only action. A review-and-referral agent is triggered by an invoice-paid-in-full event or a daily scan of completed work; a per-trade playbook supplies the ask timing, and suppression rules cap requests at two attempts with a 48-hour minimum gap, logging the reason for every suppression. Every complaint type is flagged for a human; none auto-resolves. A composite sequence is designed to wait 48 hours before the review ask and seven days before the referral touch; that sequence has not yet run.

ProofDesign notes + diagram

Safety layers for unattended operation

119 tests2 adversarial review rounds

Two safety layers for unattended operation.

More

A content pipeline where a scheduled writer drafts and self-scores short posts, then submits them to a deterministic code gate (blocklist, banned claims, a stat whitelist with inline sourcing, no personal attribution, a quality floor) before anything publishes; a one-call kill switch returns it to human approval. Beneath it, a separate outbound guard for email: two hardcoded invariants no configuration can weaken (one forbids contact with a restricted party), HMAC clearances the transport demands, an append-only ledger, and a post-send sentinel. 119 tests; two independent adversarial review rounds found 25 issues before deployment.

ProofTest excerpt (29 of 119)Design notes

Verification and evals

14/14 daily smoke18-trade weekly regression

Verification I trust more than reports.

More

A green test suite that shares a predicate with the code cannot falsify it, so I delete each gate in turn and watch which tests still pass; on the email guard that found four sub-checks that could be removed with zero failures, and they now assert behaviour. A daily smoke check fires each agent's main write and asserts a real row landed, not a success flag. A weekly voice harness replays scripted callers across 18 trades with an independent judge, recalibrated once after it scored a safety-first emergency redirect as a failure. Done means an external effect a separate checker verified.

ProofFleet smoke harnessVoice regression harness

05Customer operator / sales story

Years inside the workflow before automating it

I know this work from the inside.

More

About six years in customer-facing technical roles in heating and cooling, as a comfort advisor and project manager: discovery with homeowners and business owners, heating-load and system design, financing conversations, permits, installation coordination, technician ride-alongs and inspections.

Hundreds of sales conversations a year, each one ending in a system that had to be designed, permitted, installed and inspected. That is where the judgment in my projects comes from: I know which step gets skipped, which call gets missed, and what a customer needs to hear before they say yes.

06How I build

Deterministic where the input is structured. A model only where judgment is needed. Humans at the one-way doors.

Coming soon.

07AI in real life

Reducing the friction between uncertainty and action

The same method outside work.

More

I use my own judgment plus AI in every area where the friction between uncertainty and action is high, and the mechanism is always the same: the real documents and the real question go in, a structured checklist or comparison comes out, and every line is read by me before anything is filed, chosen or acted on. The model does not decide and does not file.

Market monitor and scheduled accumulation

A personal market-monitoring and scheduled-accumulation agent: eleven scheduled jobs on a small Linux server, running since April 2026.

More

Deterministic scripts score a five-component market regime daily from free public data, apply a long-term trend filter to a watchlist, snapshot a paper portfolio, and write a graded weekly review to my phone. Anything touching real money executes only after an approval reply from me; an unanswered request expires after two hours and places nothing. A file-based kill switch is checked at every point on the order path, and the self-heal loop may restart monitoring jobs but never the money path.

ProofWrite-up

Life admin: taxes, benefits, first-time parent

Personal administration, the same way: a tax-season paperwork pass, a benefits-enrollment comparison, and the research that comes with being a first-time parent.

More

Source documents and the actual question in; a checklist, a comparison table, or a list of questions to ask a professional out; every line reviewed by me before anything is filed or chosen. The value is not the generated text. It is shortening the distance between an unfamiliar problem and a reviewed, concrete next step.

ProofWrite-up

Health routine and vegetarian meal planning

Health and diet as a running project: a standing assistant that holds my routine and plans a vegetarian week around it, adjusting when travel or work breaks the pattern.

More

Nothing clinical, no numbers to chase; just a consistent plan that survives contact with a real schedule.

ProofWrite-up

08Personal AI operating system

A private second brain, an MCP layer, spoken briefings

A private second brain I built for my own work.

More

Postgres with pgvector holds verbatim evidence chunks and extracted memories, every row carrying provenance and a sensitivity level. Local open-weight models on one small server embed and extract nightly under a CPU cap, and a Markdown wiki is regenerated from the database each night. Eleven MCP tools expose search and recall to my coding assistant behind a per-client sensitivity ceiling. A publication firewall, a claims ledger, an approval queue and a deterministic leak gate wired into the site build, is the only path from private notes to this site. A spoken-briefing path to my phone is prototyped.

ProofSource: site, firewall, assistant, Jarvis

09Hobbies

Off the clock

Coming soon.

10Contact

Email is the door

Email is the door: [email protected]. If you are building or deploying AI where real customers and real operations are involved, I would like to hear about it.

[email protected]

Talk to my second brain

Prefer to ask instead of read? The assistant below answers questions about my work using only the text on this page: what a project does, why it matters, what I bring to a role. It will not discuss my personal life or how anything was built; for that, write to me.

Answers come only from the text on this page. Personal questions and build details get a polite no and my email: [email protected].