XPMeXperience · Pharma Manufacturing
Engineering Intelligence for Pharma Manufacturing
Intelligent software that connects experience and data on the pharmaceutical manufacturing floor.
We start with coating. XPM Golden Recipe compares each run against recipe-specific Golden Profiles; XPM Troubleshooting organizes defect investigation into causes, checks, and reusable cases — with engineers making the final call.
Visual simulation · not process data
Products
Two products. One coating process.
One works with measured run data, the other with what people observe on the floor. Both exist to support an engineer’s review.
- Product 01 · Run data comparison
XPM Golden Recipe
“Where did this run differ from the reference?”
Compare coating runs against recipe-specific reference profiles to surface differences worth reviewing
- PDF run report ingestion
- Recipe-specific Golden Profiles
- Top 5 differences for special batches
- Patterns across flagged batches
- Product 02 · Defect case review
XPM Troubleshooting
“What should we check first?”
Organize coating-defect investigation with rule-based cause candidates, practical checks, and reusable cases
- Rule-based cause candidates with evidence
- Verification checklist and possible responses
- LOT cases and similar cases
- Template drafts and external AI prompts
At a glance
Overview
Processing flow
- 1PDF run reports
- 21-minute normalization
- 3Process features
- 4Recipe reference
- 5Exception comparison
Product temp. · DEMO-BATCH-K vs reference
Sample curveReference batches by recipe
| Recipe | Reference | Status |
|---|---|---|
| DEMO-RCP-01 | 14 | Comparable |
| DEMO-RCP-02 | 9 | Comparable |
| DEMO-RCP-03 | 6 | Comparable |
| DEMO-RCP-04 | 3 | Few samples |
Overview
Cases per month
Plain counts · not a defect rateLatest review
DEMO-LOT-024Sticking · Twinning · Mid-run
- 1Over-wetting / over-sprayLikely11
- 2Insufficient dryingLikely9
- 3Higher actual spray ratePossible6
Sum of rule matches · not a probability
Recently registered
| LOT | Defect | Top candidate |
|---|---|---|
| DEMO-LOT-024 | Sticking | Over-wetting / over-spray |
| DEMO-LOT-023 | Color non-uniformity | Suspension non-uniformity |
| DEMO-LOT-022 | Logo bridging | Viscosity / solids / leveling |
| DEMO-LOT-021 | Abrasion | Weak core tablets |
Why we exist
Coating issues rarely have a single cause. The evidence is already on the floor — in run reports, in batch records, in people who have seen it before. We build tools that bring that evidence together, and leave the judgment to the people who make it.
- 01
Run data is locked in reports
Each batch leaves a PDF run report. You can open them one by one, but lining a batch up against other runs of the same recipe is hard.
- 02
Know-how stays with people
Knowing what to check first lives with senior operators, so newer team members repeat the same trial and error.
- 03
Records don’t feed the next decision
Cases and reports exist, but they are hard to search by condition — and hard to link to how a similar batch was resolved.
Product 01
XPM Golden Recipe
Compare coating runs against recipe-specific reference profiles to surface differences worth reviewing
XPM Golden Recipe reads the PDF run reports produced by coating equipment, normalizes them into 1-minute process data, and builds an observed reference — a Golden Profile — from normal batches of the same recipe. It shows where a batch’s trends and process features departed from that reference, so engineers can narrow down what to look at.
From run report to engineering review
PDF run reports
Ingest equipment run reports and record parsing status and warnings.
Normalized process data
Align 10-second and 1-minute sampling to a 1-minute basis; derive process features and data-quality info.
Recipe reference
Build an observed Golden Profile from normal batches sharing the same recipe key.
Exception comparison
Compare a special batch with the reference distribution and list the largest feature differences.
Engineer review
A difference is a place to look, not a cause. People make the call.
XPM Golden Recipe preview
Walk through the main screens, from trend comparison to data quality. Every recipe, batch, and curve shown is fictional.
A batch curve over the reference band
Overlay a batch’s product, inlet, and exhaust temperatures, pump speed, or chamber pressure on the median and range of its recipe reference.
- Trends normalized to 1 minute
- Process phases marked
- Out-of-band segments highlighted
Trend comparison
Product temp. (°C)
DEMO-RCP-01 · 1 minSynthetic curves on a 1-minute basis. Differences mark what to review — they are not a cause determination.
Batch process features
| Feature | Batch | Ref. median |
|---|---|---|
| Total run time | 132 min | 121 min |
| Min. product temp. (coating) | 36.9 °C | 40.4 °C |
| Drying phase duration | 19 min | 15 min |
| Mean inlet temp. (coating) | 58.6 °C | 58.0 °C |
| Spray pauses | 2 | 1 |
An observed reference per recipe
Summarize medians and distributions of process features from normal batches, and show where a batch sits within them.
- Reference statistics per recipe key
- Batch position within the distribution
- Not an optimal recipe or acceptance limit
Golden Profile
DEMO-RCP-01 · 14 reference batches
| Process feature | Median | P10–P90 | DEMO-BATCH-K position |
|---|---|---|---|
| Total run time min | 121 | 116 – 127 | 132 |
| Mean product temp. (coating) °C | 41.6 | 40.9 – 42.3 | 40.1 |
| Mean inlet temp. (coating) °C | 58 | 56.8 – 59.4 | 58.6 |
| Max. pump speed rpm | 18 | 17 – 19 | 19.5 |
| Mean chamber pressure diff. Pa | -82 | -90 – -75 | -84 |
| Drying phase duration min | 15 | 13 – 17 | 19 |
A flagged batch vs. its reference
Compare a batch flagged for yield or appearance issues with normal batches of the same recipe, and list its top 5 feature differences.
- Approximate percentile vs. reference
- Reference sample size and caveats shown
- Differences are review indicators, not causes
Special batch review
Flagged batch
Appearance issue note (DEMO)
Comparison baseline
- Recipe key
- DEMO-RCP-01
- Reference batches
- 11 (same recipe, marked normal)
- Excluded
- 3 flagged batches
Top 5 feature differences
vs. reference distribution · approx. percentile- 1Min. product temp. (coating)36.9 °CReference 40.4 °C ≈ P5
- 2Drying phase duration19 minReference 15 min ≈ P95
- 3Max. pump speed19.5 rpmReference 18.0 rpm ≈ P90
- 4Total run time132 minReference 121 min ≈ P90
- 5Exhaust temp. range4.8 °CReference 2.9 °C ≈ P85
With few reference batches, rankings shift easily. Differences shown are indicators for further engineering review, not causes.
Differences that repeat across flagged batches
See which differences recur in the same direction across several flagged batches, next to external-factor notes entered by your team.
- Direction matrix by feature
- Sorted by recurrence
- Manual notes alongside
Group patterns
Group patterns
4 flagged batches · DEMO-RCP-01| K | M | Q | T | Repeats | |
|---|---|---|---|---|---|
| Total run time | Above reference | Above reference | Above reference | Above reference | 4/4 |
| Drying phase duration | Above reference | Above reference | Above reference | Within range | 3/4 |
| Min. product temp. (coating) | Below reference | Below reference | Within range | Below reference | 3/4 |
| Max. pump speed | Above reference | Within range | Above reference | Within range | 2/4 |
| Mean chamber pressure diff. | Within range | Within range | Below reference | Within range | 1/4 |
Features that differ in the same direction in at least 3 of 4 batches are listed first for review.
External-factor notes (manual)
- DEMO-BATCH-KSuspension prep delayed (note)
- DEMO-BATCH-M—
- DEMO-BATCH-QRight after maintenance
- DEMO-BATCH-TRaw material LOT change (note)
What was read, and how
Check parsing status, sampling normalization, warnings, and event logs per report, and produce a text summary report.
- Parsing history and warnings
- Event logs
- Text summary report draft
Reports · data quality
Reports · data quality
| Report | Batch | Sampling | Status |
|---|---|---|---|
| DEMO-REPORT-048.pdf | DEMO-BATCH-T | 10 s → 1 min | Parsed |
| DEMO-REPORT-047.pdf | DEMO-BATCH-S | 1 min | Warning3-minute event gap |
| DEMO-REPORT-046.pdf | DEMO-BATCH-R | 10 s → 1 min | Parsed |
| DEMO-REPORT-045.pdf | DEMO-BATCH-Q | 1 min | WarningSome header fields unread |
Event log · DEMO-BATCH-K
- 00:10Spraying started
- 00:58Spray paused (2 min)
- 01:14Spray paused (1 min)
- 01:35Switched to drying
- 02:12Run ended
Summary report draft
Compared with the DEMO-RCP-01 reference, DEMO-BATCH-K ran longer overall and in the drying phase, and its product temperature stayed below the reference band during part of the mid-coating phase. These differences are flagged for further review.
Key capabilities
PDF run report ingestion
Parse coating equipment run reports and track report metadata, parsing status, and warnings.
1-minute normalization
Align 10-second and 1-minute sampling to a 1-minute basis and produce derived trends and a data-quality report.
Batch trends and features
Review batch-level trends and features such as temperatures, pump speed, chamber pressure, and phase durations.
Recipe Golden Profiles
Build observed reference statistics from normal batches with the same recipe key and compare operating patterns.
Special batch comparison
Compare manually flagged batches with reference batches and show the top 5 differences and distribution positions.
Group patterns and summaries
Summarize recurring differences across flagged batches, event logs, and text summary reports.
What it does
- Compares batches using run reports — showing differences from an observed, recipe-specific reference
- Warns when the reference sample size is small
- Lets your team flag special batches and add external-factor notes by hand
- Runs locally on a PC
What it does not do
- Act as an AI that calculates optimal recipes or recommends settings
- Determine defect causes automatically, or make quality or release decisions
- Control equipment in real time or connect to MES / SCADA
- Provide approved acceptance criteria or validated limits
Want to talk about XPM Golden Recipe?
We welcome thoughts on how run reports are used and how reference baselines should work.
Product 02
XPM Troubleshooting
Organize coating-defect investigation with rule-based cause candidates, practical checks, and reusable cases
Enter the defect type and process conditions, and a set of maintained rules with weighted scores ranks candidate causes, shows the evidence behind each, and lists priority checks and possible responses. Reviewed cases are stored per LOT and reused for similar-case lookup and occurrence tracking.
From first check to a reusable record
Capture defect and conditions
Enter the observed defect, when it appeared, and the state of process, suspension, equipment, and core tablets.
- Defect · timing · severity
- Tablet / exhaust / inlet temperature
- Prep timing · mixing · foam · nozzle
Review causes and evidence
Conditions that match rules add up to a score per cause. Top candidates appear with the evidence behind them.
- Rule match → score per cause
- Likely · possible · needs checking
- Contributing conditions shown
Organize what to check
Priority checks, possible responses, and deviation review points are laid out like a checklist for the floor.
- Priority checks
- Immediate response candidates
- Deviation review points
Record and reuse
Saved as a LOT case, it feeds similar-case suggestions, occurrence counts, and draft wording.
- Register and search LOT cases
- Up to 5 similar cases
- Template-based drafts
XPM Troubleshooting preview
Walk through the main screens in order. Every LOT, equipment name, and number shown is fictional sample data.
Inputs → candidate causes → checks
Inputs sit on the left, ranked candidate causes with their evidence in the middle, and priority checks and possible responses on the right.
- Scores are sums of rule matches, not probabilities
- Contributing inputs shown for each cause
- Final judgment stays with your team and QA
Cause review
Inputs
Case
- Defect
- Sticking, Twinning
- Timing
- Mid-run
- Severity
- Moderate
Process
- Tablet temp.
- Low
- Exhaust temp.
- Low
- Pump speed
- High
- Wetness
- Over-wet
- Tablet flow
- Tablets sticking
Suspension
- Prepared
- Previous day
- Mixing
- Insufficient
- Foam
- Moderate
Equipment · spray
- Pump hose
- Suspected wear
- Nozzle
- Normal
- Spray pattern
- Normal
Core tablets
- Hardness
- Normal
- Chips / cracks
- None
Candidate causes
Sum of rule matches · not a probability- 1Over-wetting / over-sprayLikelyScore11
- Over-wet state entered
- Tablets sticking together
- High pump speed
- 2Insufficient dryingLikelyScore9
- Low tablet temperature
- Low exhaust temperature
- 3Higher actual spray ratePossibleScore6
- High pump speed
- Suspected hose wear
- 4Suspension non-uniformityNeeds checkingScore4
- Prepared the previous day
- Insufficient mixing
- 5Pump / hose issueNeeds checkingScore3
- Suspected hose wear
Priority checks
- Compare actual tablet and exhaust temperature trends with the previous LOT
- Check the gap between set pump speed and actual spray rate
- Inspect the pump hose for compression, wear, and seating
- Confirm mixing and settling of the suspension before use
Possible responses
- Pause spraying and check whether drying recovers
- Re-measure spray rate against the set range
- Confirm the suspension after re-mixing
Deviation review points
- Review over-wetting onset together with spray volume and actual rate
- Confirm whether appearance issues decreased after the response
Find LOT cases by condition
Filter cases by equipment, defect type, timing, and suspected cause — and see which past cases share conditions with the selected one.
- Similar cases ranked by weighted condition match
- Shared conditions shown explicitly
- CSV export
Case library
Case library · 6
| LOT | Process date | Equipment | Timing | Top candidate | Deviation |
|---|---|---|---|---|---|
| DEMO-LOT-024 | 2026-06-21 | DEMO-COATER-B | Mid | Over-wetting / over-spray | Review |
| DEMO-LOT-019 | 2026-05-12 | DEMO-COATER-A | Early | Insufficient drying | Not needed |
| DEMO-LOT-016 | 2026-04-27 | DEMO-COATER-B | Late | Tablet shape / logo area | Review |
| DEMO-LOT-011 | 2026-04-02 | DEMO-COATER-B | Mid | Over-wetting / over-spray | Required |
| DEMO-LOT-009 | 2026-03-18 | DEMO-COATER-C | Mid | Higher actual spray rate | Not needed |
| DEMO-LOT-006 | 2026-02-20 | DEMO-COATER-B | Early | Insufficient drying | Not needed |
Similar cases
For DEMO-LOT-024Weighted condition match · up to 5
- DEMO-LOT-01117pts
Improved after response
- DEMO-LOT-00612pts
Improved after response
- DEMO-LOT-0199pts
Under review
What, where, and how often
Compare case counts by month, equipment, defect type, and cause category on one screen.
- Plain occurrence counts
- Never converted into defect or improvement rates
- Multi-selected defects counted individually
Occurrences
Cases per month
casesCases by equipment
casesCases by defect type
casesCases by cause category
casesA case can record several defect types and causes, so totals can differ from the number of cases.
A faster first paragraph for review documents
Seven templates are filled from the saved case. A prompt for external AI tools can be generated with masking options.
- Template-based drafts — not automated judgment
- QA review statement included by default
- The app never calls an AI API itself
Drafts
Draft type
- Troubleshooting summary
- Deviation background
- Cause analysis
- Impact check points
- Action summary
- Recurrence prevention
- Training text
- External AI prompt
The rule-based scoring identified over-wetting / over-spray, insufficient drying, and a higher actual spray rate as the causes to review first.
This is an assessment of possibilities linked to the entered conditions — low tablet temperature, low exhaust temperature, high pump speed, an over-wet state, and tablets sticking together — and does not establish a single cause.
Points to examine include the relationship between over-wetting onset and spray volume, the actual tablet and exhaust temperature trends, and the gap between set pump speed and actual spray rate.
This text is a reference draft. Final root cause, product quality impact, CAPA adequacy, deviation classification, and batch release require review by the responsible staff and QA.
External AI prompt masking
- Product name
- Batch number
- Equipment
- Date
The app never calls an AI API. It only produces text for you to copy.
Key capabilities
Structured condition entry
Record defect type, timing, process conditions, coating suspension, spray equipment, and core tablet status in consistent fields.
Candidate causes with evidence
Rules and weighted scores rank possible causes, and each candidate shows which inputs contributed to its score.
Checks and possible responses
Priority checks, immediate response candidates, and deviation review points linked to the top causes.
LOT cases and similar cases
Register, search, edit, and export cases to CSV; see up to five past cases ranked by condition match.
Occurrence overview
Case counts by month, equipment, defect type, and cause category, side by side.
Drafts and external AI prompts
Seven template-based drafts, plus a prompt you can paste into an external AI tool yourself, with masking options.
What it does
- Organizes candidate causes from maintained rules and scores — shown as qualitative levels (likely · possible · needs checking)
- Lets you maintain rules, scores, and equipment / defect / cause master data to fit your site
- CSV / JSON backup and restore of cases and reference data (admin area)
- Runs locally on a PC — data stays in a local file
What it does not do
- Decide the final root cause, product quality impact, CAPA, deviation class, or batch release
- Replace batch records, deviation management, or any other GMP system
- Calculate statistical cause probabilities or defect rates, make ML predictions, or analyze images
- Call AI APIs directly, or connect to MES / SCADA equipment data in real time
Want to talk about XPM Troubleshooting?
We welcome thoughts on review practices, rule design, and case management.
How they fit
What the data shows,and what the floor should check.
The two products answer different questions in a coating investigation. Used side by side, they let you review measured run differences next to floor observations.
Where did the measured run differ?
Compares a batch with reference batches of the same recipe to find differences in trends and process features.
- Based on run reports
- Observed reference per recipe
- Largest feature and time-segment differences
Which causes and checks should we review?
Organizes cause candidates, checks, and past cases from the observed defect and conditions.
- Based on observations and inputs
- Rule-based cause candidates
- Checklists and cases
Today the two products run independently. They do not share a database or exchange data automatically — the flow above describes how a person can review both results together.
Approach
Visible evidence. Human judgment.
Software for manufacturing should be as clear about what it does not know as about what it does.
Evidence is never hidden
Which data and which rules produced a result is always visible.
People make the call
Root cause, quality impact, deviation, CAPA, and release decisions stay with your team and QA.
Local first
Current versions run on a PC and keep data in your local environment.
No inflated claims
We don’t publish unverified accuracy, probability, or impact figures.
Technical profile of the current versions
XPM Golden Recipe
- Runtime
- Runs locally (Python · Streamlit)
- Input
- Equipment PDF run reports
- Reference
- Observed statistics per recipe (Golden Profile)
- AI integration
- None
XPM Troubleshooting
- Runtime
- Runs locally (Python · Streamlit · SQLite)
- Cause review logic
- Rules + weighted scores
- Similar cases
- Weighted condition match (up to 5)
- AI integration
- None — copyable prompt only
About
Built from experience on the manufacturing floor
XPM — eXperience · Pharma Manufacturing — grew out of hands-on work in oral solid dose manufacturing. We build software that connects what experienced people know with the data processes already produce.
Our first domain is coating. We respect the language and procedures of the floor: our tools support engineering review alongside existing quality systems, rather than replacing them.
XPM Labs
XPM Labs is our development and research brand — where prototypes and new process tools take shape, starting with the next steps after coating: tableting and granulation.
Products and processes
- Products
- XPM Golden Recipe · XPM Troubleshooting
- Process focus
- Coating (film coating of oral solid doses)
- Planned
- Tableting · Granulation
Contact
We’d like to hear from the floor.
Thoughts on using coating run data, reviewing defect cases, potential fit, or collaboration are welcome.