Internal Automation
hiring, sales outreach and LinkedIn, run by agents with a human in the loop
- Client
- Ainoviq, Blackngreen's AI division
- Role
- Full-stack and AI Engineer
- Duration
- ongoing since 2026
- Delivered
- 1 May 2026
- Status
- live
The problem
A lot of company time went on work that is structured but not quite mechanical: screening and coordinating candidates, researching and writing outreach, keeping LinkedIn activity going. Each is too repetitive to enjoy and too judgement-dependent for a simple rule engine — which is exactly the shape of problem language models are good at, provided somebody stays in the loop.
Constraints
- Anything that sends on someone's behalf needs a human approving it. An agent that emails a candidate or a prospect unsupervised is a reputational risk, not a productivity win
- Outreach that reads as generated defeats the purpose, so the writing had to be specific enough to be worth receiving
- Hiring automation touches candidate data, so what gets stored and inferred is a design constraint rather than an afterthought
- These run against real platforms with real rate limits and real terms of service
What I did
- Automated the hiring process end to end, from intake through coordination, with the judgement calls surfaced rather than hidden
- Built Meridian, a human-in-the-loop LinkedIn and email outreach platform — leads, campaigns, pipeline and inbox, where every message is drafted for a person to send rather than sent by the agent
- Automated LinkedIn activity within the same human-approval model
- Kept the approval gate as the product, not a setting — the value is in the drafting, and the human is what makes it safe to use
The hard part
Making the draft good enough to actually send. A human-in-the-loop system only saves time if the human is approving rather than rewriting — and the first version of anything like this produces text that is faster to bin than to fix. Getting past that meant giving the agent real context to work from and being ruthless about the tells: no opening flattery, no invented common ground, no sentence that could have been sent to anyone.
Outcome
The automations run across hiring, sales outreach and LinkedIn, and the outreach work has produced real conversions.
What this did not do
- The conversion figures are internal and are not mine to publish, so this page claims the outcome without a number attached to it. Take it as unverified from the outside.
- Nothing here sends autonomously. That is a deliberate design decision and also a limit — it does not scale past what a person will approve.
- Draft quality is uneven across segments. It is strongest where there is real context to work from and weakest where there is not.
- This is current work at my employer; the systems, clients and metrics stay confidential.
Stack
Related results
Results
- viditparashar.meviditparashar.me › experience › ainoviq
Full-stack and AI Engineer at Ainoviq (Blackngreen) (Mar 2026 – present)
Full-timeMar 2026 – present
1 Mar 2026 — Building AI products inside Blackngreen's AI division and running several at once — an AI companion on a full-duplex model, and the automation that now handles hiring,…
- viditparashar.meviditparashar.me › projects › clinical-agents
Clinical Agent Suite — four AI agents for the parts of care that are…
Case studyShipped2025
31 Aug 2025 — … pilots, not asserted. Designed and built four separate agents — medical scribe, patient intake, clinical decision support, and medical coding — on Next.js, LangChain…
- 40%physician documentation time saved
- 4clinical agents designed and shipped
- 2academic health systems integrated
- Next.js
- TypeScript
- Python
- LangChain
- LLMs
- Node.js
- +2 more
- viditparashar.meviditparashar.me › projects › resup
Resup.dev — a résumé builder aimed at the software that reads it first
Case studyLive2026
31 Mar 2026 — … résumé and produces a document a parser can read, with templates and export options. Engineered the résumé parsing and optimisation pipelines in Python using NLP and…
- 2input formats accepted
- parse → structure → optimise → exportpipeline stages
- unknownmeasured effect on interview rate
- Next.js
- TypeScript
- Python
- Tailwind CSS
- NLP
- LLMs