Shreyansh Panda AI consulting & automation systems Book a call

Your team is doing what your systems should be.

I find where your operations lose time, money and attention, then design and build the AI and automation systems that take that work off your team.

Book a call How I work

Taking new work for Q4

I don’t start with AI. I start with the constraint.

Most automation fails because it was built on a process nobody audited. I map how work actually moves through your business: where it queues, where it gets re-entered by hand, where one person waits on another. Then I fix that.

Sometimes the answer is an AI agent. Sometimes it’s deleting three steps and connecting two tools. The honest recommendation is worth more than the impressive one.

Eight things, one thread running through them.

  • AI Consulting

    Where AI genuinely pays for itself in your business. And, just as usefully, where it doesn’t.

  • Workflow & Process Audits

    A map of how work moves today, with a time and money cost attached to every handoff.

  • Business Automation

    Removing the manual steps sitting between your tools, your data and your team.

  • AI Systems

    LLM-powered systems built around your data, your rules and your actual edge cases.

  • Agentic Workflows

    Agents that carry out multi-step work, with review points exactly where they matter.

  • Operational Optimization

    Fewer steps, shorter cycle times, less re-entry, cleaner handoffs between people.

  • Research & Discovery

    Structured investigation of the problem before anybody writes a line of code.

  • Implementation

    I build the system, ship it, and hand it over documented, not as a slide deck.

Five stages. No stage gets skipped.

Every engagement starts with a problem stated in business terms, not technical ones. AI and automation are chosen last, and only where they genuinely beat the simpler option.

  1. 01

    Discover

    Understand the business, the workflow, the bottleneck and the outcome you actually want. I ask about revenue, headcount and cycle time before I ask about tools.

    • Stakeholder interviews
    • Constraint brief
    • Success metrics
  2. 02

    Audit

    Map the current process end to end and quantify where time, money, friction and manual work are being lost. Most teams have never seen this on a single page.

    • Process map
    • Time & cost model
    • Bottleneck ranking
  3. 03

    Design

    Decide what actually solves it: AI, automation, a software change, a process change, or some combination. The recommendation always includes what not to build.

    • System architecture
    • Build vs. buy
    • Phased roadmap
  4. 04

    Build

    Implement with the tools that fit the job: LLM workflows, agents, n8n, APIs, Python, your existing stack. Shipped, tested and documented, not left as a prototype.

    • Working system
    • Documentation
    • Team handover
  5. 05

    Optimize

    Measure against the baseline from the audit and improve it. A system nobody monitors quietly degrades until people route around it.

    • Monitoring
    • Iteration
    • Quarterly review

Discovery informs the audit. The audit decides the design. Nothing gets built on a guess.

The stack, arranged the way work actually flows.

  1. 01 Input

    Intake

    Work arrives from wherever it already lives.

    • Forms
    • Email & calendar
    • WhatsApp
    • CRM events

  2. 02 Understanding

    Enrich

    Each item is researched, completed and checked before anything acts on it.

    • Lead research & enrichment
    • Retrieval & RAG
    • Validation & reconciliation

  3. 03 Decision

    Decide

    Rules and models choose the route. A person reviews where it matters.

    • LLM workflows
    • AI agents
    • Evaluation & guardrails
    • Human-in-the-loop

  4. 04 Execution

    Act

    The system does the work between your tools.

    • APIs & webhooks
    • CRM automation
    • Document pipelines
    • Queues & retries

  5. 05 Output

    Report

    Every run is logged, measured and fed back into the next.

    • Reporting & analytics
    • Dashboards

    monitor · evaluate · improve

Intelligence
  • AI models
  • LLM workflows
  • AI agents
  • Retrieval & RAG
  • Evaluation & guardrails
  • Prompt systems
Orchestration
  • n8n
  • Workflow orchestration
  • Event triggers
  • Queues & retries
  • Human-in-the-loop
  • Scheduling
Integration
  • APIs & webhooks
  • CRM automation
  • WhatsApp automation
  • Email & calendar
  • Internal business systems
  • Document pipelines
Data
  • Python
  • Data workflows
  • Lead research & enrichment
  • Validation & reconciliation
  • Reporting & analytics
  • Dashboards

Tools are chosen last. The architecture is chosen first, and it survives a change of vendor.

Problems, constraints, and what the system changed.

B2B services agency · 34 people

Cutting a six-day onboarding to the same afternoon.

Problem
New client onboarding ran six working days across five tools and forty-plus manual steps.
Constraint
No budget for new headcount, and the ops lead was a single point of failure for every account.
System
An orchestrated intake pipeline: form, enrichment, document generation, CRM records, workspace provisioning, with one human approval gate.
Implementation
n8n orchestration, LLM document drafting, CRM and workspace APIs, Slack approvals, audit logging.
Onboarding time
6d → 4h
Returned per month
110 hrs
Missed setup steps
0

Specialist recruitment firm

Research that took a day, ready before the morning call.

Problem
Consultants spent three to four hours per role manually researching companies and building shortlists.
Constraint
Accuracy mattered more than speed. One confidently wrong output would have killed adoption.
System
An agentic research workflow that gathers, verifies and scores candidates against a role brief, with a citation behind every claim.
Implementation
Python pipelines, LLM agents with tool use, enrichment APIs, a separate verification pass, CRM write-back.
Per role
3.5h → 20m
Consultant capacity
+40%
Cited output
100%

D2C brand · high message volume

Answering seven in ten support messages without a person reading them.

Problem
Twelve hundred WhatsApp messages a week, and most of them were the same eight questions.
Constraint
Brand voice was non-negotiable, and anything touching refunds had to stay with a human.
System
A retrieval-backed WhatsApp assistant with a hard scope boundary, confidence thresholds and clean escalation to the team.
Implementation
WhatsApp Business API, retrieval over policy and order data, guardrails, tiered handoff, resolution analytics.
Auto-resolved
71%
First response
4h → 2m
Escalation tiers
3

Professional services · 12 partners

A monthly reporting pack that builds itself.

Problem
Client reporting consumed two people for a full week every month, almost entirely copy-paste.
Constraint
Every number had to reconcile exactly with the finance system. Close enough was not acceptable.
System
A data workflow that pulls from four systems, reconciles them against source of truth, then drafts narrative commentary for partner review.
Implementation
Python ETL, validation rules, LLM commentary drafting, templated output, a review queue with sign-off.
Per month
70h → 6h
Systems reconciled
4
Reporting errors
0

What the people who hired me said afterwards.

He spent the first two weeks asking questions instead of building. That is exactly why the thing he built actually gets used.
Operations Director, B2B services agency
We had been quoted for a platform. He showed us we needed three automations and one process change. It cost a fraction of that and it worked.
Founder, recruitment firm
The audit on its own was worth the engagement. We had never seen the whole process on one page with a cost attached to each step.
Head of Delivery, D2C brand

I build systems that take work off people.

I work with businesses and agencies that have outgrown their own processes. Teams where the workflow was designed for five people and now carries twenty-five. Where a spreadsheet quietly became infrastructure. Where the best person on the team spends their week on data entry.

My background is in systems: how work moves, where it stalls, and what technology can honestly fix. I use AI where it earns its place: document work, research, classification, drafting, multi-step decisions with a review point. Where a simple automation or a deleted step does the job, I say that instead.

I run the discovery, the audit, the design and the build. That combination is the whole point. A recommendation from someone who will never have to implement it is a guess.

Have a process that shouldn’t require humans to do this much?

Tell me what’s slow. I’ll tell you whether it’s worth automating. And if it isn’t, I’ll tell you that too.

Book a call

30 minutes. No pitch deck. Bring one process you would like to stop doing by hand.

Or write to shreyanshpanda@proton.me