Selected Projects

Work that eliminates the manual layer

A selection of my own prototypes, internal tools and demos in AI automation, from compliance tracking to operational intelligence. Each one starts from a real process problem, not a technology brief. These are not paid client projects.

Research prototype 01 / 04

Research prototype: wildlife image review

Independent project · Computer vision

The Problem
Camera trap projects produce thousands of images per season, and sorting them by hand is slow and uses specialist time that is hard to find at scale. This is the problem the prototype explores.
What Was Built
An independent prototype I built from scratch to explore how vision models can pre-sort field images and route uncertain ones to a person. Images are ingested, classified, confidence-scored and filed into structured reports; low-confidence captures are flagged for human review rather than discarded.
What it explores
Less review The idea is to cut the time spent reviewing images. Clear identifications are filed automatically; low-confidence captures go to a person to check. In my own testing on real field images, about 90% were filed automatically and only the uncertain ones went to a person. It is a prototype, not a production system or a client project.

Approach

A multi-stage pipeline I designed and built myself, using the concept behind Google vision models: image ingestion → AI classification → confidence filtering → structured output generation. Human review is kept for low-confidence results, so oversight stays in the loop.

Key Components

  • Automated image ingestion and pre-processing
  • AI-powered species classification
  • Confidence scoring with human escalation for edge cases
  • Structured report generation
Live — Internal 02 / 04

AI Business Development Pipeline

Sales Automation & Outreach Intelligence

The Problem
Finding, researching, and writing personalised outreach to prospects across several industries — without a sales team. Researching each firm by hand takes a long time, which adds up quickly across a few hundred firms.
What Was Built
A seven-stage pipeline: niche research → prospect identification → ICP scoring → AI-personalised email generation → Gmail draft creation → sequenced follow-up → reply tracking. Drafts are held for my review; nothing is sent without a yes.
Outcome
140+ personalised outreach emails drafted across two niches in under two weeks, held as drafts for review before anything is sent.

Approach

Built as a modular pipeline where each stage hands off to the next. AI handles prospect research, company context extraction, and email personalisation. Gmail's API creates drafts in bulk for human review before sending — keeping a human in the loop on the final action.

Key Components

  • Automated niche and prospect research
  • ICP scoring against fit criteria
  • GPT-powered email personalisation per firm
  • Gmail API bulk draft creation
  • Sequenced follow-up with reply detection
  • Bounce and deliverability monitoring
In Development 03 / 04

Cybersecurity Intelligence Platform

Security Operations & Threat Monitoring

The Problem
Security teams need to monitor, correlate, and act on threat intelligence from multiple sources simultaneously. Manual triage is slow and inconsistent — threats get missed, and the volume of noise makes prioritisation difficult.
What Is Being Built
A real-time intelligence platform that aggregates threat feeds, correlates signals across sources, surfaces prioritised alerts, and generates structured incident reports — reducing analyst time spent on triage and repetitive documentation.
Target Outcome
∞→1 Multiple fragmented threat feeds unified into a single prioritised intelligence view. Analyst attention directed to decisions, not data aggregation.

Approach

Built on a modern full-stack foundation (Next.js, Supabase, Clerk) with real-time data ingestion, automated correlation logic, and structured alert generation. Designed for extensibility — new threat sources can be added without rebuilding the core.

Key Components

  • Multi-source threat feed aggregation
  • Automated signal correlation and deduplication
  • Prioritised alert surfacing
  • Structured incident report generation
  • Role-based access and audit trail
Service overview 04 / 04

Business Workflow Automation

Operations & Process Automation — SMEs

The Problem
UK SMEs losing hours each week to repetitive manual tasks: copying data between systems, chasing leads that go cold, generating reports by hand, and managing compliance paperwork that doesn't change — only the dates do.
What I Build
Custom automation workflows designed around each business's existing tools. Lead routing, CRM updates, document generation, scheduled reporting, email follow-up sequences — built to run without anyone watching them.
How a Build Runs
One process scoped first, built and tested on your own data, then handed over with a written note. How much time it saves is measured in the audit, not promised in advance.

Workflows I Can Build

  • Lead qualification and CRM routing from web forms
  • Automated follow-up email sequences with reply detection
  • AI-drafted email responses to inbound enquiries, held for review — never auto-sent
  • Document and report generation from structured data
  • Compliance checklist tracking and evidence collection
  • Invoice and onboarding document automation
  • Cross-system data sync (removing copy-paste work)

How It Works

An engagement can start with a free 45-minute discovery call that checks fit. Automation is then scoped to the highest-value process first, and other workflows follow once the first is working. No large upfront commitment required.

Demo — AI-Drafted Reply

Two AI-drafted replies, screenshotted directly from a live Gmail account. Held for review before anything sends — illustrative demo scenarios, not real client correspondence.

Have a process worth automating?

Book a free 45-minute discovery call. We check whether I am the right fit. I give no advice on the call.

Book a free discovery call
Sarat Sarvepalli

Sarat Sarvepalli

AI automation consultant

13 years building production software across healthcare, government, and enterprise — now applied to AI automation and workflow systems for UK businesses.

LinkedIn →