Question 1

What is Mithtech's automation expertise stack — Polygin + n8n + GraphQL + API engineering?

Accepted Answer

Mithtech's automation engineering capability is built on a five-layer stack — picked deliberately so we can deliver bespoke automation across any client architecture. The five layers, in the order they typically appear in client engagements:

1. **[Polygin](https://polyg.in/)** — Mithtech's first-party WhatsApp + customer-comms SaaS. Polygin handles WhatsApp Cloud API broadcasts + templates + visual flow builder + multi-agent inbox + AI chatbot (OpenAI / Gemini / DeepSeek) + ERPNext / Frappe CRM auto-sync. Hosted on Mithtech's Oracle Cloud infrastructure with Indian data residency. Anchor for any automation engagement that has a WhatsApp / customer-conversational dimension. Pairs with **[Polygin Integration](/products/polygin-integration)** (the MIT-licensed Frappe app at github.com/suparikoli/polygin) when ERPNext-side custom DocType extension is needed.

2. **[n8n](/products/n8n)** — open-source workflow orchestration. Self-hosted (Apache-2.0), 400+ native integrations, AI Agent node, custom Mithtech-built nodes. The integration backbone tying everything else together — webhook receivers, scheduled jobs, conditional routing, multi-step pipelines, error-handling, retry strategies.

3. **GraphQL + REST API engineering** — the layer that makes everything else possible. Mithtech's API craft includes:
    - **GraphQL** — Apollo Server / Yoga / Hasura / Pothos / Nexus / TypeGraphQL schema design, federation across multi-service architectures, subscriptions for real-time data, persisted queries for performance, schema evolution patterns.
    - **REST** — OpenAPI 3 schema-first design, JSON:API conventions, versioning strategies, HATEOAS where it earns its keep, consistent error-response shapes.
    - **Webhooks** — idempotent receivers with signature verification, retry-with-exponential-backoff, dead-letter queues, replay tooling for debugging.
    - **Event streams** — Kafka / Redis Streams / NATS / AWS EventBridge consumption with consumer groups + offset management.
    - **OAuth / OIDC** — integration with Auth0 / Keycloak / Clerk / Workspace OAuth / Microsoft identity / Sign-In-with-Apple.
    - **Rate limiting + observability** — token-bucket / sliding-window strategies, distributed Redis-backed rate limiters, OpenTelemetry tracing, p99 latency dashboards, error-budget tracking.

4. **Workflow automation patterns + production engineering** — beyond just "use n8n", Mithtech brings a production-pattern library refined across deployments: idempotent webhook receivers, retry + exponential backoff + dead-letter queues, circuit breakers around flaky third-party APIs, distributed locks for race-condition-sensitive flows, audit logging that satisfies DPDP / GDPR / HIPAA-equivalent requirements, secret rotation without workflow downtime, A/B-testable prompt versioning for LLM-driven flows, evaluation harnesses with golden test sets.

5. **AI / LLM layer** — Botpress Cloud (when LLMz is needed) + direct OpenAI / Anthropic / Groq / Hugging Face / Ollama integration (for embedded AI features) + custom RAG pipelines (Postgres pgvector / Qdrant / Meilisearch / Typesense).

**Why this stack matters for the client:** any automation engagement above a certain complexity threshold needs all five layers in some combination. A client thinking "I just need to connect WhatsApp to my CRM" is buying a one-layer engagement (Polygin + Polygin Integration). A client thinking "automate our entire B2B sales-to-fulfillment pipeline" is buying a five-layer engagement (Polygin for WhatsApp touchpoints, n8n for cross-system flows, GraphQL/REST for connecting to their existing apps, production patterns for reliability at scale, AI/LLM for the smart parts like email-to-order extraction). Mithtech's value is bringing the right combination per client — not over-engineering small projects or under-engineering large ones.

Question 2

How is automation-ai different from just using n8n / Botpress / Polygin Integration directly?

Accepted Answer

Those are **tools**. automation-ai is the **engineering service that uses those tools** (and custom code) to deliver real production automation. Same shape as the difference between buying a Volkswagen Polo and hiring a transportation logistics company — one is a vehicle; the other is the operation that uses vehicles to solve a problem.

**When to engage at the tool level:**
- **[/products/n8n](/products/n8n)** — "Install n8n on our infrastructure and help us set up the first 5 workflows" (₹3–8 L scope, 4–8 weeks). Turnkey install + initial workflows + handover. Client's team takes ongoing ownership.
- **[/products/polygin-integration](/products/polygin-integration)** — "We need WhatsApp ↔ ERPNext bridge for templated broadcasts and order updates" (₹80k–2.5 L scope). Free MIT app + paid Mithtech configuration.
- **[/products/botpress](/products/botpress)** — "Build us one specific Botpress Cloud agent for customer support" (₹2–6 L scope). Agent design + Code Hooks + deployment. Client's team operates ongoing.

**When to engage at the service level (automation-ai):**
- **"We have 12 manual processes across 5 systems and want a phased roadmap to automate them strategically"** — discovery + architecture + multi-tool implementation + ongoing iteration (₹6–30 L+ scope, 3–12 month programme).
- **"We want AI features in our SaaS product but don't have ML expertise on our team"** — Mithtech engineers join as fractional ML/automation team, build evaluation harnesses, ship features incrementally.
- **"We're not sure which tool fits — n8n vs Botpress vs build custom — and want someone to architect the right choice per use case"** — discovery-led engagement; tool selection becomes part of the design.

Most mature client engagements graduate from a single-tool start ("install n8n") to the service-level engagement ("now help us think strategically about everything we should automate"). automation-ai is the home for that broader, ongoing relationship.

Question 3

What does Mithtech actually build vs configure off-the-shelf?

Accepted Answer

We build **custom code** when the off-the-shelf tools don't reach. Distribution from production deployments:

**Off-the-shelf tool configuration (~50% of project hours):**
- n8n workflows authored in the n8n UI but version-controlled via Mithtech-built export-to-Git tooling
- Botpress agents designed in Agent Studio with custom prompts and tool-call configuration
- Polygin Integration configured for client's specific WABA templates and trigger events
- Frappe Server Scripts and Workflows for ERPNext / Frappe CRM / Frappe Helpdesk side
- Direct LLM API calls (OpenAI / Anthropic / Groq) wired into client app codebases

**Custom code we build (~50% of project hours):**
- **Custom n8n nodes in TypeScript** — for client systems where the 400+ native integrations don't reach (SAP, Tally, proprietary APIs, legacy databases). Published as private packages installed into the client's n8n instance.
- **Custom Botpress integrations via @botpress/cli** — Mithtech-built bridges to ERPNext, Frappe CRM, Polygin, Smartflo, internal client systems.
- **RAG pipelines** — document ingestion + chunking + embedding + vector store (Postgres pgvector / Qdrant / Meilisearch / Typesense) + retrieval + reranking + LLM call. Custom evaluation harnesses to measure retrieval quality on client-specific corpora.
- **Prompt engineering + evals** — version-controlled prompt libraries with golden test sets, regression alerts when LLM provider model upgrades change behaviour.
- **Workflow orchestrators in Python / TypeScript** — when the use case is genuinely complex enough that an n8n graph becomes unmaintainable (typically multi-tenant SaaS automation features, or workflows with deep concurrent fan-out).
- **Mithtech-built observability dashboards** — Grafana panels consuming n8n execution metrics + Prometheus app metrics + LLM-call cost telemetry + business-outcome KPIs.
- **Cost-control middleware** — wrapper around LLM calls that caches, batches, falls back to cheaper models when latency permits, kills runaway costs at configurable thresholds.

All custom code lives in the **Mithtech-owned Git repository** delivered to the client. Client owns the IP. We follow standard engineering hygiene — code review, tests, CI/CD, documentation, runbook.

Question 4

What does this typically cost — and where does the cost go?

Accepted Answer

**Implementation cost** (one-time, paid to Mithtech): ranges widely with scope.

| Engagement shape | Typical scope | Cost |
|---|---|---|
| **Focused pilot** | One specific automation (e.g., email-to-order parsing for a B2B distributor) — discovery + build + 4 weeks of tuning | ₹2–4 L |
| **Multi-workflow programme** | 5–10 workflows across 2–3 systems + observability + handover | ₹6–12 L |
| **AI-feature programme for SaaS** | Embedded AI features (smart suggestions / RAG / summarisation / classifier) inside the client's product, with evals + A/B testing | ₹8–18 L |
| **Multi-phase enterprise programme** | Full discovery + 3–4 phases of build over 6–12 months + ongoing iteration retainer | ₹15–30 L+ |

**Where Mithtech's hours go (typical 12-week programme breakdown):**
- Discovery + architecture: ~15%
- Workflow design + build: ~30%
- Custom code (nodes / integrations / RAG pipelines): ~30%
- Testing + evals + observability setup: ~10%
- Documentation + handover + training: ~10%
- Iteration buffer (post-launch tuning): ~5%

**Ongoing platform costs** (paid by client directly to upstream vendors — Mithtech does NOT resell or mark up):
- **n8n** — self-hosted is free; n8n Cloud is $20–800+/month depending on workflow count + execution volume. Mithtech recommendation: self-hosted via [Coolify](/products/coolify) on Mithtech-managed infrastructure.
- **Botpress Cloud** — Plus $79/mo, Team $445/mo, Managed $1,245/mo + AI spend (see [/products/botpress](/products/botpress) for the full pricing table).
- **LLM providers** — OpenAI ($0.15/1M input tokens for GPT-4o-mini up to $5/1M for GPT-4o), Anthropic (Claude 3.5 Haiku $0.25/1M to Claude Sonnet 4 $3/1M), Groq (very cheap for Mixtral / Llama-3 inference), Hugging Face Inference Endpoints (for fine-tuned models). Mithtech tunes per-step model selection during the first month to minimise this line item.
- **Vector DB / observability** — typically self-hosted via Coolify (Postgres pgvector / Qdrant / Grafana / Prometheus / Loki) so no incremental SaaS fees.

**Ongoing Mithtech retainer** (optional): ₹1–5 L/month for iteration + ops + new-feature development as the client's automation footprint grows. Most clients keep us on retainer for at least 6–12 months post-launch.

**Honest disclosure on cost variance:** automation-ai is a custom-engineering engagement and final cost depends heavily on the depth of integration (off-the-shelf nodes vs custom code), the maturity of the client's systems (clean APIs vs legacy databases requiring bridging), and the desired observability + governance level. We scope these honestly during discovery — no surprises at month 12.

Question 5

When does Mithtech recommend automation-ai over hiring an in-house automation engineer?

Accepted Answer

Honest answer: **for sustained internal automation operations, hiring a senior in-house automation engineer is often the better long-term answer** — at the right scale. A senior automation/ML engineer in India costs ₹30–80 L/year fully loaded. Below that scale (or before that scale is justified), Mithtech is the right call. **Mithtech's decision framework:**

**Hire in-house when:**
- The automation footprint is large enough to keep one engineer fully utilised (typically 15+ active workflows + ongoing new-feature pipeline + monitoring/maintenance)
- The use cases are deeply tied to the client's core IP and stay in flux (rapid iteration cycles where contractor handoff overhead would slow things down)
- The client has the engineering management bandwidth to interview, onboard, and retain ML/automation talent
- Budget supports the ₹30–80 L/year fully-loaded cost

**Engage Mithtech when:**
- The client doesn't have ML/automation expertise yet and needs to ramp from zero — Mithtech architecting + initial build + knowledge transfer is faster than 6+ months of in-house hiring → ramp.
- The pipeline of automation work is bursty (large initial scope then steady-state maintenance) — fractional Mithtech retainer is more economical than one full FTE
- The client wants outside expertise validating their architecture and prompt engineering choices
- The use case requires specialised skills the client doesn't want to build in-house permanently (RAG-pipeline tuning, custom n8n node development, multi-LLM cost optimisation)
- Client size is sub-200 employees — building an internal ML team usually requires 250+ employee scale to justify
- The client wants to start, validate ROI, and *then* hire in-house for ongoing ops once the system is proven

**Common pattern:** Mithtech does the first 3–6 months (discovery + build + initial deployment + 4 weeks of tuning), then a senior in-house hire takes over ongoing ops with Mithtech on retainer for monthly architecture reviews and complex new features. We're transparent about this trajectory — if the right answer for the client at month 12 is to bring it in-house, we help with the transition rather than holding the relationship hostage.

Question 6

Can automation-ai work alongside the client's existing internal IT team?

Accepted Answer

Yes — that's the most common shape. Most Mithtech automation-ai engagements run alongside a client-side internal team (IT ops, software engineering, data engineering). **Standard collaboration patterns:**

1. **Shared Git repository** — Mithtech-owned at engagement start, transitions to co-owned (client-side reviewers required for production merges) by month 3, fully client-owned by handover. Mithtech engineers continue committing as external contractors with PR-based workflow.
2. **Shared observability** — Grafana dashboards accessible to both teams; alerts route to a shared Slack/Mattermost channel; on-call rotation arranged per SLA tier.
3. **Knowledge transfer cadence** — weekly architecture reviews during the build phase; monthly retro + roadmap during operate phase; complete documentation handover at any milestone (not just at programme end).
4. **Boundary clarity** — Mithtech owns the automation engineering surface; client team owns the systems being automated (ERPNext / CRM / databases / app code). When automation needs touch into core client codebases, work is done as PR contributions to the client's repo, not in Mithtech-private branches.
5. **No vendor lock-in** — every artifact is portable. n8n workflows export to JSON; Botpress projects export via the SDK; custom code is in standard TypeScript / Python with documented dependencies. The client's team can take over without Mithtech, hire a different contractor, or move to a different automation stack — all on their schedule.
6. **Hybrid retainer models** — some clients keep Mithtech on a 20-hour/month or 40-hour/month retainer for complex new features while their in-house team handles routine ops. Others have Mithtech do all the engineering and the in-house team manages the business-side configuration.

We explicitly position automation-ai as an **augmentation of the client's existing team**, not a replacement. The clients who get the most value are those where Mithtech and the in-house team work well together — we screen for that during discovery.

Automation & AI Agents — Mithtech Engineering Service

What can Automation & AI Agents do?

Mithtech automation expertise stack — Polygin (first-party WhatsApp SaaS) + n8n (workflow orchestration) + GraphQL + REST API engineering + workflow automation patterns + AI/LLM layer

How does Automation & AI Agents work?

Audit & scope

Deploy on your infrastructure

Customise & integrate

Train & roll out

Operate & evolve

Automation & AI Agents vs the alternatives

Estimate Automation & AI Agents ops cost at your volume.

Frequently asked questions about Automation & AI Agents

Get Automation & AI Agents running in weeks, not quarters.

Products that pair with Automation & AI Agents — Mithtech Engineering Service

n8n

Polygin Integration for Frappe

AIOS — AI Sales Intelligence Platform

Chatwoot

ERPNext

Medusa B2B — Wholesale Commerce on Medusa.js

Know your budget? Plan your Automation & AI Agents — Mithtech Engineering Service stack.