Analytics / AI / Automation
I'm Vignesh. I build systems that turn structured and unstructured data into business decisions that create value. I also build software to automate processes, so you can focus on your core expertise and deliver more value to your clients and business.
Explore the workTwo domains.
One systems mindset.
Accounting & Finance
Knowledge visualization
A searchable statute index becomes a branching network. Find a section, follow its path through the knowledge structure, and arrive at the exact heading.
Try it: search for 80C, then choose a result.
Adapted from my work for Gundesha & Associates. An index visualization, not tax advice.
Document intelligence
Watch a statement move through page alignment, row detection, field extraction and balance verification before becoming structured accounting entries.
Illustrative workflow with synthetic values.
Healthcare
Built on the visual logic of my perioperative analytics dashboards: compare booked time with what actually happens, see the gaps, and understand where the day runs long.
DE-IDENTIFIED PDF PATTERNOperating room performance
Same cases / 30-minute turnovers / additional room
Perioperative flow / inspect the same room
Black outline: scheduled patient in to scheduled patient out. Color: actual patient in to patient out. Internal milestones are illustrative; room turnaround is the actual gap between cases.
Case durations are preserved, with 30 minutes between every pair of cases. Both arrangements use the same comparison window, shown above. Dashed slots include capacity already available before rearrangement; the before/after difference drives the revenue change. Fewer occupied rooms can still mean fewer usable slots for a particular case duration. Moving cases requires compatible staffing, equipment and patient readiness. Revenue is an estimate from your input, before costs.
Capacity over time
Compare the de-identified source pattern with three synthetic days designed to illustrate rooms becoming available.
Same operating rooms and 10-hour daily capacity.
Case duration accuracy
The selected day's cases grouped by actual duration versus scheduled duration.
Duration categories follow the dashboard's 20- and 40-minute thresholds.
Estimate before the bill arrives
Separate the charges that stay fixed from equipment and disposables that vary. Each dot is one bill in the same procedure class.
Colored dots highlight the current step. Grey dots keep the other bills in view.
PatientPlan around a range, and see which items could move the final bill.
HospitalSeparate the repeatable base from equipment and consumable variation.
InsurerAssess bill variability before applying actual coverage terms and patient share.
This demonstration generates 500 independent synthetic cases in one procedure class. Fixed charges are INR 75,000 per case. Equipment and disposables vary; no real patient observations or identifying source records are used.
The narrow gold interval is a 95% percentile bootstrap confidence interval for the average (400 resamples). It is not the range for an individual patient's bill. The wider teal range is a 90% split-conformal prediction interval: the first 80% estimate a baseline mean, and the remaining 20% calibrate absolute errors using the finite-sample rank. Coverage assumes future cases follow the same exchangeable process. This has not been validated on real hospital bills.
More cases can improve the estimate of the average without eliminating variation between patients. Ranges can widen or narrow as cases arrive. Dots stack within bill-amount bands; vertical spacing rescales per view. Shading marks the central share of observed simulated amounts, not a prediction guarantee. Equipment/consumable amounts here are bill components, not verified costs or margins.
Billing precision
Follow the charges from service records to a reconciled bill, then see what the patient, hospital and insurer each need to understand.
Package charge: INR 1,50,000. Supported extras: INR 30,000. Non-covered extras: INR 10,000. A duplicate INR 5,000 is removed before allocation. Reconciled total: INR 1,90,000.
Illustrative policy: INR 1,80,000 eligible, INR 10,000 deductible, then 10% co-pay on the remaining INR 1,70,000. Patient: 10,000 + 17,000 + 10,000 = INR 37,000. Insurer: INR 1,53,000. Combined: INR 1,90,000. Non-covered items remain valid bill items; a duplicate is removed.
Inspired by package-versus-actual and item-level billing analytics. All amounts, review findings and policy terms in this public walkthrough are illustrative. No original patient, doctor, hospital, insurer or claim identifiers are reproduced.
The original work connected utilization, case-length accuracy and day-level room timelines. The first view reconstructs 17 visible room lanes from a dashboard PDF, with approximate minute timings and all original names, dates and case identifiers removed. Scheduled and actual bars are paired by left-to-right order; duration comparisons in this view are illustrative, since case identity cannot be verified from the export. Three additional days use synthetic cases to demonstrate capacity changes. These are illustrative scenarios, not measured savings or clinical recommendations.
Start a conversation
Analytics, AI & automation engineering