AI agents & chatbots
Website and WhatsApp assistants that answer only from your documents, qualify leads, and hand over to a human. LLM agents with approval steps and capped spend.
Taking new projects · replies within 24h
We automate the boring part of your business.
Fixed price. Shipped in days.
A small senior engineering team from India — AI agents & chatbots, n8n and Python automations, backend & database work, and Android apps. Every price is on this page. Every package has a written definition of done. You own the code.
- 5.0★ from 10 reviews on Freelancer.com since 2015
- ~10 yrs lead-level engineering on large production systems
- From ₹1,999 / $29 · INR & USD pricing
Prices on the pageNo “contact us for pricing”.
Definition of doneAgreed in writing before we start.
Milestone payments50% to start, 50% on approval.
You own itYour accounts, your repo, no lock-in.
01 · Services
Four things we’re genuinely good at.
We say no to everything else (see what we won’t do).
Automation workflows
One repetitive job that just runs: forms to Sheets or CRM, inbox to summaries, PDFs to spreadsheets, daily reports, review and listing data pulls.
Backend & data
FastAPI and REST APIs, scheduled data pipelines on GCP, PostgreSQL tuning and load testing, and fixes for broken scripts.
Android apps
Crash and bug fixes, new screens and features in your existing app, small MVP apps and Play Store releases.
02 · What automation actually changes
Same lead. Two very different evenings.
03 · Packages & prices
Small, fixed-price packages.
Pick one, or use the estimator below. Click any card to see exactly what “done” means.
Automation
₹1,999Async automation audit
Not sure what to automate first? We look at how you work and tell you — no call needed.
Definition of done
- A written report or 5-min Loom: your top 3 automations
- Rough time saved and what each would cost
- A fixed quote for automation #1
- 100% credited if you buy a package within 14 days
You send: A short description or screen recording of your current process
Not included: Any build work
Automation
₹3,999Listings & reviews pull
Google (and Yelp for US) ratings, review counts and links for your list of locations, in one clean sheet.
Definition of done
- Clean Google Sheet / Excel with rating, count and listing link
- Unmatched or closed locations flagged, never guessed
- Official APIs where the platform requires it
- Up to 1,000 locations: ₹9,999 / $149
You send: Your list of names + addresses
Not included: Personal data, logins to third-party sites
Automation
₹4,999One workflow, end to end
One repetitive job that just runs: form → Sheet/CRM, inbox → summary, PDF → Sheet, daily report.
Definition of done
- Built in your n8n, Make, Zapier or Google account
- Up to 4 apps and ~12 steps, with error alerts
- Optional AI step (classify, summarise, draft reply)
- Handover notes + short video, 14 days of fixes
You send: Which apps, an example input and the result you want
Not included: Paid app subscriptions or AI usage (billed to your account)
AI bots
₹7,999Website FAQ + lead-capture bot
A chat widget that answers only from your own pages, says “I don’t know” instead of guessing, and captures qualified leads.
Definition of done
- Answers cite the page they came from
- 3–5 qualifying questions → Google Sheet + email alert
- Staging link to test before it touches your site
- Embed snippet, 20-question test sheet, Loom handover
You send: Your website URL or docs, and the questions you want asked
Not included: CRM / booking hookup (add-on), AI usage after 30 days (moves to your key)
AI bots
₹9,999WhatsApp FAQ + lead capture
An official WhatsApp Business assistant that answers FAQs, qualifies leads and hands over to a human.
Definition of done
- Official WhatsApp Business Platform (no grey-market tools)
- FAQs, qualifying questions, human handover
- Every chat logged to a Google Sheet
- English by default; Hindi as an add-on
You send: Your FAQs, a business number, and Meta Business access
Not included: Meta message fees (paid to Meta), go-live waits on Meta verification
Backend & data
₹2,499Python script fix
A broken script or small automation, fixed — with the root cause explained in plain words.
Definition of done
- Root cause + fix + short change note
- Clear errors instead of silent failures
- +₹2,500 / $50: logging, config file, 3–5 unit tests
You send: The script, a sample input, the exact error
Not included: Access to your production servers
Backend & data
₹9,999PostgreSQL performance check
Slow queries, bloated tables, connection storms — find out why, with evidence, before you buy a bigger server.
Definition of done
- Top slow queries with EXPLAIN ANALYZE evidence
- Index, wide-column and connection-pool findings
- Ranked fix list with expected impact
- Safe, reversible migration scripts for the top fixes
You send: Read-only access to a staging copy or pg_stat_statements export
Not included: Running changes on production without your DBA’s sign-off
Android
₹3,999Android crash or bug fix
A crash, ANR or broken screen in your Android app, found and fixed.
Definition of done
- Reproduced, fixed and explained
- Tested on emulator + one physical device
- Pull request or patched source you own
You send: Repo access and steps to reproduce (or a crash log)
Not included: iOS, Play Store policy appeals
Android
₹9,999Android feature or screen
One new screen or feature in your existing Android app — login, form, list, payments hand-off, push notifications.
Definition of done
- Kotlin code following your app’s existing patterns
- Wired to your API, with loading and error states
- Signed test build (APK/AAB) to try before merge
You send: Repo access, a design or sketch, the API docs
Not included: Backend changes (quote separately), store listing design
Android
₹39,999Android MVP app
A small native Android app to test an idea: up to 5 screens, a simple backend, and a Play Store release.
Definition of done
- Up to 5 screens, Kotlin, your branding
- Simple backend (Firebase or our API) if needed
- Play Store internal-testing release, source in your repo
- Paid in 3 milestones
You send: Your idea in a page, a rough sketch of the screens
Not included: iOS, complex offline sync, payment-gateway certification
Care plan
Content refresh, break-fix and one small change a month. Cancel anytime. Never required.
Prices as listed, in INR or USD. On Upwork and Freelancer, platform prices apply. Third-party costs (AI usage, Meta WhatsApp fees, app subscriptions) are billed to your own account at cost — never marked up.
04 · Estimator
Price it yourself in 20 seconds.
No email gate. This is the same maths we use for quotes.
05 · Try a mini demo
A chatbot that admits what it doesn’t know.
This one is scripted and runs entirely in your browser — no AI calls, no data sent anywhere. The bots we build for you work the same way on your documents: answers cite their source, and anything off-topic gets “I don’t know — here’s how to reach a human”.
- Grounded only in your pages and files
- Shows where each answer came from
- Captures the lead and alerts you
Fixed Scope assistant scripted demo
06 · Work
What we’ve built — labelled honestly.
Real work = shipped in production at our jobs (described generically, no employer code). Demo = built to show the pattern, with sample data. Open source = public repo you can read. Private project = our own tool; the code isn’t public, happy to walk you through it.
AI coding-agent migration pipeline
- Problem
- Moving dozens of production APIs from a legacy service to Go by hand took 2–3 days per endpoint and risked behaviour drift.
- What we built
- A Python-orchestrated coding-agent pipeline: read-only tracing of each endpoint, generated idiomatic Go, mandatory human review and parity checks, per-route traffic splitting for safe rollout.
- Result
- Most of 40 migrated endpoints went from 2–3 days to under half a day — roughly 80% less effort.
Spec-to-tasks agent planning harness
- Problem
- LLM agents turning product specs into epics and tasks burned tokens on raw documents and lost control on long runs.
- What we built
- A staged pipeline that builds a verified, source-cited context first, then decomposes level by level with propose → human approve → write gates, ledgers and resumable runs.
- Result
- Auditable, resumable agent runs with a human approval at every level.
FAQ chatbot that refuses to guess
- Problem
- Support bots that invent answers are worse than no bot.
- What we built
- A RAG assistant over a fictional coffee shop’s docs: local retrieval, citations for every answer, and an explicit “I don’t know” below a confidence threshold.
- Result
- Public demo you can run without an API key (mock mode).
MCP server toolkit
- Problem
- Letting an AI assistant touch your data and APIs safely needs guard rails.
- What we built
- A Python Model Context Protocol server: GitHub tools, read-only SQL, OpenAPI-driven HTTP calls, blocked internal-network requests, tests and CI.
- Result
- Works with Claude Desktop and Cursor over stdio or HTTP.
RFC-to-plan Claude Code plugin
- Problem
- Approved design docs rarely turn into a plan engineers can execute task by task.
- What we built
- A four-stage plugin that resolves open decisions, drafts a repo-grounded PLAN.md with file:line references and test tables, then verifies coverage against the RFC.
- Result
- Asks a human only when a real decision is needed.
Google Sheets cleanup + weekly report
- Problem
- A sales sheet full of duplicates, bad dates and nobody summarising it.
- What we built
- Apps Script automation: trims, de-duplicates and validates rows, moves bad rows to their own tab with reasons, builds a week-over-week summary on a schedule.
- Result
- Ready-to-copy sheet with unit-tested logic.
Website to spreadsheet
- Problem
- Someone copy-pastes product data from a website into Excel every week.
- What we built
- A polite scraper over a public practice site that collects 1,000 products across 50 pages into clean CSV and Excel.
- Result
- Sample output included in the repo.
Config-driven partner integrations
- Problem
- Every new subscription-benefits partner needed its own custom integration.
- What we built
- One standard activation flow: an orchestrator API, partner config that picks the path, events on Kafka — instant, scheduled, stacked activations.
- Result
- For most new partners, onboarding became a config change instead of new code.
PostgreSQL latency fix at high traffic
- Problem
- A hot read path regressed to ~120 ms p99; the obvious index didn’t help because one value made up ~99.9% of rows.
- What we built
- Confirmed with EXPLAIN ANALYZE, replaced it with a partial index on the rare, hot values; traced cost to wide jsonb projections and connection hold time.
- Result
- p99 went from ~120 ms to ~50 ms without indexing rows nobody queried.
PostgreSQL load testing & tuning
- Problem
- Will the database hold at peak, and what breaks first?
- What we built
- Load tests covering fat columns, skewed indexes, buffer memory, index read:write ratio and connection-pool sizing.
- Result
- Capacity limits and tuning changes backed by measurements, not guesses.
Serverless logistics platform on AWS
- Problem
- A new cross-border logistics business had to validate workflows fast, without waiting on servers.
- What we built
- Lambda + DynamoDB, a Step Functions shipment lifecycle with retries, carrier webhooks normalised into one status model, SQS/SNS, infra as code with CDK.
- Result
- Shipped in weeks and scaled with the business without a re-architecture.
Scheduled data pipeline on GCP
- Problem
- Data from several APIs needed to land clean, on time, every day.
- What we built
- Python jobs on Cloud Run + Cloud Scheduler with data-quality checks (missing, duplicate, stale), PostgreSQL with versioned migrations, secrets in Secret Manager, Slack alerts on failure.
- Result
- Runs unattended; failures page a human instead of failing silently.
Service metrics auditor
- Problem
- Dashboards that look fine until the incident you didn’t instrument.
- What we built
- A Claude Code plugin that compares what a service emits against its dashboard, names the blind spots, and writes an improved dashboard plus runbooks.
- Result
- Backend-agnostic: Prometheus/Grafana, Datadog, OpenTelemetry.
Native Android apps
- Problem
- Small teams with an Android app and no in-house Android developer.
- What we built
- New screens and features, crash and ANR fixes, API integration, and Play Store releases — in Kotlin or Java, following your app’s existing patterns.
- Result
- Case studies will be added here as we ship Android work.
07 · How we use AI
Yes, we use AI. Here’s exactly how.
A senior engineer reviews every line. Here’s how we work with these tools.
- Assistants may draft. Coding tools can draft boilerplate, tests and docs. A senior engineer decides what ships.
- A senior engineer reviews every line. Nothing ships that one of us hasn’t read, run and tested.
- Your data stays yours. We don’t paste private data into AI tools without your written OK, and we use your own API keys for production.
- Bots don’t guess. Our assistants answer from your sources and say “I don’t know” otherwise. Medical, legal and financial advice are switched off.
- Spend is capped. AI usage runs on your account with limits set, so a bug can’t become a bill.
- You own the code. Full source, workflows and credentials handed over. No licence fees to us.
08 · What we won’t do
Saying no is part of the service.
No fake reviews or inflated metrics.
Everything on this page is checkable. If it’s a demo, it says demo.
No “AI will run your business”.
We automate one job well. Big vague projects get split into a first fixed slice.
No grey-area scraping or spam.
Official APIs, public data, platform rules. No bulk WhatsApp blasts or bought lists.
No trading or betting bots.
Not now, not for any budget.
No surprise invoices.
Fixed price agreed up front. Scope changes are re-quoted before work starts.
No overpromising timelines.
If something depends on Meta, Google or your team, we say so up front.
09 · Process
From brief to done in five steps.
- Day 0Send a brief
Two lines, a screen recording or a doc. No call needed.
- < 24hFixed quote
Price, timeline and the definition of done, in writing.
- StartFirst milestone
Usually 50%. UPI / bank in India, platform escrow abroad.
- Day 1–5Staging preview
You test it on real examples before it touches anything live.
- HandoverDone = done
Final payment on approval. Source, notes, video — and 14 days of free fixes.
10 · Who you’ll work with
Two engineers. Zero account managers.
You talk directly to the person writing your code.
Tushar Gupta
AI automation & backend engineer
Lead engineer, ~10 years building large-scale consumer-tech systems. 5.0★ from 10 reviews on Freelancer.com since 2015.
Builds the AI agents, automations and backends. Day job: leading engineering on high-traffic production systems. Evenings: shipping small, well-scoped projects for teams that agencies ignore.
PythonLLM agents & RAGMCP toolsn8n / Make / ZapierFastAPI & REST APIsPostgreSQL tuningGCP & AWSData pipelinesWeb data extraction
Calls are weekdays 6 to 9 PM IST, booked 24 hours ahead. Async replies within 24 hours.
11 · FAQ
Straight answers.
Why are your prices so much lower than agencies?
No account managers, no office, no sales team — two senior engineers, reusable templates from past work, and AI to speed up the typing. We keep packages small on purpose so we can price them honestly.
Is a call required?
No. Send a two-line brief or a screen recording and we reply with a fixed quote within 24 hours. If you prefer a call, we take them on weekdays from 6 to 9 PM IST, booked 24 hours ahead (that's morning on the US East Coast, early afternoon in the UK).
How do payments work?
Fixed price, split into milestones: usually 50% to start and 50% when you approve the delivery. Indian clients pay by UPI or bank transfer; international clients can hire us through Upwork or Freelancer for escrow protection.
Who owns the code and the workflows?
You do, from the day you pay. Everything is built in your accounts or handed over as source code — no lock-in, no licence fees to us.
What counts as “done”?
Each package lists its definition of done. We agree it in writing before we start. If the delivery doesn’t match it, we fix it — and every package includes 14 days of free fixes after handover.
What if my project doesn’t fit a package?
Use the estimator for a ballpark, then send a brief. Bigger projects get split into a first fixed-price slice so you can judge us on something small.
Do you work with clients outside India?
Yes. We’re India-based and price in both INR and USD. Working hours overlap with US mornings and UK afternoons, and most work is async anyway.
Are AI model costs included?
The build is. Ongoing AI and WhatsApp/Meta usage is billed to your own account, so you see the real cost and nothing is marked up. We set spend caps where the provider supports them.
12 · Contact
Tell us the boring job. We’ll quote it in 24 hours.
Fixed price, written definition of done, no obligation. Prefer a booking page? Message or book on Topmate.
- No call needed — async by default
- Calls, if you want one: weekdays 6 to 9 PM IST, booked 24 hours ahead
- We never share your details or add you to lists