Failures are reported
A script that fails silently is more dangerous than the manual task it replaced. Every failure sends a notification to a person.
Freelance Python Developer in Bangalore
For the manual job somebody does every week.
Most of my Python work is not building websites. It is removing the repeated manual task – the spreadsheet somebody reconciles every Monday, the report assembled by hand, the data pulled from one system and typed into another.
- Automation and data work
- Scripts that log and retry
- Scheduled and monitored
- Replies within one working day
Overview
Where automation earns its money.
Almost every business has tasks that a person does repeatedly because nothing joins two systems together. Downloading a report, reformatting it, cross-referencing it against another file, and emailing the result. Copying orders from one platform into another. Checking a supplier’s prices by hand. Assembling the same monthly figures from four places.
These tasks are individually small and collectively enormous. Four hours a week is five working weeks a year, and the cost is not only the hours – it is that the work gets skipped when someone is on leave, and that a manual process performed two hundred times contains mistakes nobody has found.
Python suits this work particularly well. The standard library and ecosystem cover spreadsheets, PDFs, databases, HTTP, email and scheduling without much ceremony. A useful script can be a hundred lines. That means automation is worth doing for tasks far too small to justify building software in a heavier language.
The other half is data work: cleaning and reconciling messy exports, transforming data between systems that disagree about formats, generating reports from several sources, and scraping information that is only available on a web page. Unglamorous, and reliably the thing that frees up the most time.
What I offer
Pythonwork
01
Task automation
Scripts that do the repeated manual job on a schedule, with logging and alerts when something goes wrong.
Hours returned every week, and a task that happens even when someone is away.
02
Data processing & cleaning
Turning messy exports into consistent, usable data – deduplicating, normalising and validating.
Decisions made on data you can trust rather than a spreadsheet nobody quite believes.
03
Report generation
Pulling figures from several sources and producing the same report every time, on schedule.
Reports that arrive without anyone assembling them, formatted the same way each month.
04
Web scraping
Collecting data that is only published on web pages, respectfully and within what a site permits.
Information you currently gather by hand, gathered reliably instead.
05
API integrations
Connecting systems that do not talk to each other, with retries, rate limiting and proper error handling.
Data moving between systems instead of being re-keyed by a person.
06
Internal tools
Small applications and command-line tools for the jobs your team does often.
Work that currently needs a specific person can be done by anyone.
Key benefits
How scripts are written.
Unattended code has to fail loudly or it is worse than no code.
Everything is logged
What ran, when, what it processed and what it changed – so when a figure looks wrong there is a record rather than a theory.
Safe to run twice
Written so running the same job again does not duplicate data or double-count, because that will happen eventually.
Credentials handled properly
API keys and passwords in environment configuration, never committed to a repository and never in the script itself.
Anyone can run it
Clear setup instructions and sensible defaults, so it does not become one person’s private tool.
Honest about payback
If a task takes ten minutes a month, automating it will not repay the cost. You will hear that rather than a quote.
Who this is for
Who this helps.
Anyone with a recurring manual job and a person doing it.
Operations teams
Reconciling, checking and re-keying between systems that were never connected.
Finance teams
Monthly reports assembled by hand from several sources, the same way every time.
E-commerce businesses
Stock, pricing and order data moving between a store, a marketplace and an accounting system.
Agencies
Client reporting compiled manually every month across several platforms.
Businesses with legacy systems
Old software with no API, where data has to be extracted somehow.
Anyone maintaining a big spreadsheet
A workbook that has quietly become critical infrastructure.
Where I work
Python automation in Bangalore.
Bangalore is full of Python capability aimed at data science, machine learning and product engineering. What is harder to buy is a small, practical piece of automation: one task, done properly, for a business that does not want a data team.
The work here often involves Indian specifics – GST return preparation from transaction exports, reconciling payment gateway settlements against orders, and pulling data out of portals that offer a download button and nothing resembling an API.
I work with businesses across Bangalore and India, and with clients in the UK, USA, Canada, the UAE and Singapore. Most of these jobs are small, fixed pieces of work rather than projects.
Small practical jobs
One task automated properly, not a data programme.
GST and reconciliation
Return preparation and settlement matching.
Portal data
Getting data out of systems with no API.
Clients overseas
India, the UK, USA, Canada, the UAE and Singapore.
Serving Chennai, Bangalore, India, Canada, UK, USA, Dubai, Singapore.
Industries
What gets automated.
The same patterns, in very different businesses.
E-commerce
Typically needs: Stock, pricing and orders spread across a store, marketplaces and accounting.
How I help: Scheduled synchronisation with conflict rules, plus settlement reconciliation against orders.
Accounting & finance
Typically needs: Monthly returns and reports assembled by hand from several exports.
How I help: Transaction data consolidated and formatted for filing, with validation that flags anomalies.
Logistics
Typically needs: Tracking statuses checked manually across several courier portals.
How I help: Scheduled status collection into one view, with alerts on exceptions rather than a daily check.
Manufacturing
Typically needs: Production and inventory figures re-keyed between systems.
How I help: Automated extraction and loading with validation, so the numbers agree without anyone typing.
Real estate
Typically needs: Listing data maintained across several portals by hand.
How I help: One source of truth pushed out to each portal, with failures reported per destination.
Agencies & marketing
Typically needs: Client reports compiled from several platforms every month.
How I help: Scheduled collection and report generation, formatted consistently and delivered automatically.
How I work
How a job runs.
Small and fixed. Most of these are days rather than weeks.
Discovery
I watch or have described the task exactly as it is done now – every step, every exception, every judgement call somebody makes without thinking about it.
Planning
We settle what gets automated and what stays manual, where the data comes from, how often it runs and who hears about failures.
Design
The approach is designed: scheduled or triggered, where it runs, how credentials are stored, how failures are reported and what happens if it runs twice.
Development
The script is written with logging, retries and validation, and run against real data so the edge cases show up while someone is watching.
Testing & SEO
It runs alongside the manual process for a period, and the results are compared before anyone stops doing it by hand.
Launch & support
Scheduling, monitoring and alerting set up, then written instructions for running it, changing it and interpreting a failure.
Technology
What I work with.
Chosen for the task, not for the CV.
Core
Python 3Modern Python with type hints, virtual environments and pinned dependencies.
pandasData cleaning, reconciliation and transformation at useful scale.
requests & httpxAPI work with retries, timeouts and rate limiting handled properly.
openpyxl & csvReading and writing the spreadsheets the work actually arrives in.
Scraping & documents
BeautifulSoupParsing HTML where data is only available on a page.
PlaywrightSites that need a real browser, including login flows.
PDF toolingExtracting from and generating PDF documents and reports.
Running it
Cron & schedulersScheduled execution with failure alerting that reaches a person.
DatabasesMySQL and PostgreSQL where results need to be stored rather than emailed.
Logging & alertsStructured logs and notifications by email or messaging.
Features
What these jobs usually include.
Scoped against one real task rather than offered as a package.
Scheduled executionRuns unattended at the right time.
Failure alertingA person is told when something breaks.
Structured loggingA record of what ran and what changed.
Retry handlingTransient failures retried rather than abandoned.
Data validationAnomalies flagged instead of processed quietly.
Safe re-runsRunning twice does not duplicate or double-count.
Credential handlingSecrets in configuration, never in the code.
Report outputSpreadsheet, PDF or email, formatted consistently.
DocumentationHow to run it, change it and read a failure.
SEO & performance
Not a website concern.
Automation work has nothing to do with search visibility, and it would be dishonest to pretend otherwise. The relevant discipline here is different: scraping responsibly, and keeping credentials and data secure.
Where scraping is involved, I work within what a site’s terms and robots directives permit, at a request rate that does not burden it. If a site prohibits it or an API exists instead, I will say so rather than proceed.
- Scraping only where terms and robots directives permit it
- Request rates kept modest so no site is burdened
- Official APIs preferred over scraping wherever one exists
- Credentials in environment configuration, never committed
- Personal data handled only where there is a reason to
- Logs kept free of secrets and sensitive values
- Outputs stored where only the right people can read them
- A clear account of what data is collected and why
AI search
Data that is actually usable.
Most of the value in data work is making information consistent: the same field names, the same date formats, the same identifiers across systems that each had their own opinion. That is what makes data usable by anything downstream, whether a report, a dashboard or a model.
- Consistent schemas and field naming across sources
- Dates, currencies and identifiers normalised
- Validation that flags bad records instead of passing them on
- Provenance recorded – where each figure came from
- Structured output formats rather than formatted text
- Documentation of what each field actually means
Why me
Why bring it to me.
The practical reasons, rather than the adjectives.
I will tell you not to bother
If a task takes ten minutes a month, automating it costs more than it saves. You will get that answer rather than a quote, because the arithmetic is the point.
Failures are loud
Unattended code that fails silently is worse than the manual task it replaced, because at least a person notices when they have not done something. Every job alerts a human.
Small fixed jobs
Most of these are days rather than weeks, quoted as a fixed piece of work. You do not need a project or a retainer to remove one recurring task.
You are not left dependent
Written instructions, sensible defaults and no hidden setup. A script only one person can run has just moved the bottleneck rather than removed it.
FAQ
Automation questions.
Anything repetitive involving data moving between systems: downloading and reformatting reports, reconciling one export against another, copying orders or stock between platforms, generating the same monthly figures, checking prices or statuses, and collecting information published on web pages. If a person does it the same way each time, it can usually be automated.
Most of these are small fixed jobs – days rather than weeks. The cost depends on how many systems are involved and how awkward it is to get data out of them. I will also tell you the payback: if the task takes ten minutes a month, it is not worth automating and you should hear that.
It tells someone. Every job logs what it did and sends an alert on failure, with enough detail to know whether it is a transient problem that will retry or something needing attention. A script that fails silently is genuinely worse than the manual process.
It depends on the site and what you do with the data. I work within a site’s terms and robots directives, keep request rates modest, and use an official API where one exists. If a site prohibits it, I will tell you rather than proceed and leave you with the consequences.
Usually. Scheduled downloads from a portal, parsing exported files, reading PDFs, or driving the interface with a browser automation tool where there is genuinely no other route. It is less elegant than an API and it works, provided the source is not actively hostile to it.
Not necessarily – if a portal redesigns or an export format changes, a script that depends on it will need updating. That is why failures alert loudly rather than passing bad data through. Scraped sources are more fragile than APIs, and that is stated up front rather than discovered.
Depends where it runs. Scheduled jobs usually run on a server, so nobody needs anything locally. For tools your team runs on demand, I will set up the simplest arrangement that works for your environment and document it.
The data preparation, yes – consolidating transaction exports, validating them, and producing the format required for filing. The filing itself and the final responsibility stay with you and your chartered accountant, and I would rather confirm the output format with them than assume.
It runs alongside the manual process for a period and the results are compared before anyone stops doing it by hand. Discrepancies get explained rather than averaged away. Trusting automation before it has been checked is how errors get baked in permanently.
Yes. Sources change, formats change, and businesses change what they need. Most of these arrangements are occasional small changes rather than a monthly plan, and the documentation means another developer could pick it up if you would rather.