Quality Celestara

AI-Powered Product Quality & Risk Assessment

Product quality is determined by the interaction of raw materials, manufacturing processes, equipment performance, and environmental conditions — data that lives across 8+ disconnected systems at most pharmaceutical manufacturers. Current approaches, including annual CPV reviews, detect quality risks months after they emerge. Celestara changes the equation: AI agents operating on a unified quality ontology continuously assess product robustness, predict quality risks before they materialise, and deliver actionable mitigation intelligence across every product and every site.

Key Highlights

01 Consolidated quality assessment across all data sources
02 Continuous product robustness scoring for every product
03 Predictive risk identification before failures occur
04 Cross-product and cross-site risk intelligence

Regulatory Context

What FDA Inspectors Cite

21 CFR 211.100 Process Capability
1 / 6

Process validation studies approved without evaluating intra-batch or inter-batch variability in critical quality attributes

Process performance qualification studies for US-market products did not include acceptance criteria for intra-batch or inter-batch variability. PPQ batches were approved without evaluating sources of potential variation in CQAs including assay, impurities, and pH. Sampling consisted of composite results rather than enhanced sampling designed to detect variability — the firm could not demonstrate its processes were capable of consistently meeting quality standards.

Zydus Lifesciences Limited · 2024-08-29
21 CFR 211.192 Recurring OOS
2 / 6

27 out-of-specification results over five years with no holistic review to confirm the process remains validated

From May 2019 to June 2024, approximately 27 OOS results were obtained for a single API during in-process and release testing, leading to reprocessing of multiple batches. Although individual corrective actions were identified, no holistic review or effective assessment was performed to ensure the overall manufacturing process remained under adequate control. OOS investigations failed to capture critical manufacturing dates and product complaint history.

Dr. Reddy’s Laboratories Limited (Unit II) · 2024-11-19
21 CFR 211.110 Specification Justification
3 / 6

Product specification changes based on non-representative challenge studies at a fraction of commercial batch size

Hardness and thickness specifications for a tablet product were changed based on a challenge study performed on only one of multiple marketed strengths. Trial sizes evaluated were not representative of commercial batch sizes. The effectiveness check evaluated only six batches across limited strengths, with no evaluation performed for several US-marketed strengths — creating unquantified risk for products whose robustness was never assessed.

Jubilant Generics Limited · 2024-02-02
21 CFR 211.84 Supplier Quality
4 / 6

High-risk raw material components accepted on supplier certificate alone without validation or identity testing

The firm relied on supplier certificates of analysis for incoming shipments of high-risk components without establishing the reliability of the supplier’s analyses through validation at appropriate intervals. No identity or quantitation testing was performed to detect potential contaminants. Over three years, hundreds of lots of drug products containing these materials were manufactured and distributed with expiration dates of one to four years.

Catalent Indiana LLC · 2025-07-14
21 CFR 211.113 Upstream Risk
5 / 6

Eleven contamination events in upstream manufacturing over two years with unmitigated product quality risk

Upstream drug substance manufacturing systems had unmitigated risk in assurance of process performance and product quality. Evidence of eleven contamination events was documented in one building from April 2022 to February 2024, with twelve additional events in another building from January 2020 to November 2023. The firm failed to identify and address systemic deficiencies in a timely manner. This was a repeat observation from a prior FDA inspection.

Frederick Manufacturing Center · 2024-06-07
21 CFR 211.165 Statistical Controls
6 / 6

Statistical quality criteria lack defined acceptance and rejection levels — batches partially released without justification

The firm’s statistical quality control criteria failed to include appropriate acceptance and rejection levels. During batch review, approximately 85 alert excursions and 36 action-level excursions were missed in particle monitoring data. A batch was partially released without justification, and the investigation failed to account for all excursion events. Reject limits for critical defects were never established for multiple sterile products.

Jubilant HollisterStier General Partnership · 2024-06-14

The Problem

Why Product Quality Risks Go Undetected

Challenge 1 1 / 7

Quality Data Fragmented Across 8+ Enterprise Systems

  • Batch yields in MES, analytical results in LIMS, material attributes in ERP, stability in a separate database — each system holds a piece of the quality picture
  • QA teams manually export and reconcile data from 8+ systems to assess a single product’s quality posture
  • No single platform correlates a yield drift with the raw material lot, equipment change, or environmental condition that caused it
Challenge 2 2 / 7

Product Robustness Is Not Systematically Measured

  • Process capability indices (Cpk/Ppk) are calculated sporadically during validation, not maintained as continuous metrics
  • No composite robustness score integrates process stability, material sensitivity, and quality event frequency into a single product risk indicator
  • Products are considered in control until an OOS event proves otherwise — the absence of failure is mistaken for quality assurance
Challenge 3 3 / 7

Continuous Process Verification Remains Retrospective

  • Annual CPV reports compile historical data months after manufacturing occurred — too late for effective intervention
  • Statistical trending is performed in spreadsheets without connection to live process data or incoming material attributes
  • FDA inspectors increasingly expect real-time trending and proactive risk signals, not year-end data compilations
Challenge 4 4 / 7

Raw Material Variability Is Invisible to Quality Assessment

  • Supplier certificates of analysis are accepted without systematic correlation to downstream CQA performance
  • A material lot that passes incoming specifications can still cause process drift due to uncontrolled attributes outside the specification window
  • No mechanism to flag incoming materials whose variability profile predicts risk to specific products
Challenge 5 5 / 7

Quality Risks Are Detected After the Damage Is Done

  • A process drift that begins in Q1 may not trigger a CAPA until the annual review in Q4 of the following year
  • Write-offs, supply disruptions, and customer complaints are the first indicators of quality risk — not proactive alerts
  • Predictive analytics requires connected, real-time data that most quality organisations simply do not have
Challenge 6 6 / 7

Multi-Product, Multi-Site Risk Comparison Is Manual

  • Each site captures quality data in different formats, at different granularity, using different methodologies
  • Cross-product risk analysis — identifying shared equipment, materials, or process parameters that propagate risk — requires manual investigation
  • Enterprise quality leaders lack a single dashboard that ranks all products by risk and surfaces the factors driving that risk
Challenge 7 7 / 7

Rising FDA Expectations on Proactive Quality Management

  • 483 observations citing failures to assess process capability, maintain validated states, and evaluate product quality holistically are increasing
  • ICH Q12 and FDA process validation guidance require lifecycle-based quality management — from initial validation through ongoing monitoring
  • Inspectors evaluate whether firms proactively identify and mitigate quality risks, not merely react to failures after the fact

Product Quality & Risk Assessment in Celestara

1

Enterprise Data Integration and Harmonization

Celestara connects to every operational system through a purpose-built integration pipeline that normalizes heterogeneous data — structured database records, time-series sensor data, unstructured batch narratives, and supplier quality documents — into a unified data foundation. Data flows continuously, not in annual retrospective batches.

Cross-System Ingestion Schema Harmonization Data Validation Continuous Sync
2

Quality Ontology Maps Product-Process-Material Relationships

Celestara maintains a quality ontology — a connected knowledge model that maps every product to its critical quality attributes, process parameters, raw materials, equipment, and historical quality events. When a material lot changes, the ontology traces which products, processes, and CQAs could be affected. This contextual reasoning is what makes predictive quality assessment possible.

Relationship Mapping CQA-CPP Correlation Cross-Domain Querying Impact Tracing
3

Continuous Statistical Analysis and Capability Monitoring

AI agents perform continuous statistical analysis across every product — Cpk/Ppk trending, multivariate analysis, control charts — using the ontology to contextualize results. A capability drift in dissolution is automatically correlated with changes in raw material lots, equipment parameters, or environmental conditions. Analysis happens continuously, not annually.

Cpk/Ppk Trending Multivariate Analysis Control Chart Generation Anomaly Detection
4

Predictive Risk Scoring and Product Robustness Assessment

Each product receives a continuously updated robustness score derived from process capability indices, material quality signals, deviation trends, stability data, and manufacturing variability. Celestara aggregates signals across the quality ontology to identify products approaching quality boundaries before they cross them — turning reactive quality events into preventable ones.

Robustness Quantification Predictive Risk Modeling Signal Aggregation Threshold Alerting
5

Cross-Product and Cross-Site Risk Intelligence

Celestara correlates risk signals across products, sites, and supply chains. When a raw material supplier’s quality drifts, every product using that material is flagged. When a process parameter shift at one site affects quality, the system checks whether the same risk exists at other sites. Quality leaders see a single dashboard that ranks products by risk and recommends specific mitigation actions.

Cross-Product Correlation Supply Chain Risk Mapping Multi-Site Trending Mitigation Recommendation

The Solution

How Leucine Solves This

Purpose-built tools that address both the FDA compliance gaps and the operational barriers to effective product quality assessment and proactive risk management.

Celestara Quality Intelligence

Unified Quality Ontology Across All Data Sources

Celestara connects every quality-relevant system through an enterprise integration pipeline and builds a continuously updated quality ontology — a connected knowledge model that maps relationships between products, processes, materials, equipment, and quality events. AI agents equipped with domain-specific tools query this ontology to answer questions that no single system can answer alone: which material lots correlate with yield dips, which process parameters predict stability risk, and which products share equipment-related failure modes.

  • Enterprise integration pipeline connecting MES, LIMS, ERP, stability, supplier quality, and environmental monitoring into a unified data foundation
  • Quality ontology mapping product-process-material relationships with automatic impact tracing across the portfolio
  • AI agents with domain-aware tools that query cross-system data through the ontology in real time
Celestara Predictive Risk

Continuous Robustness Scoring and Predictive Risk Assessment

Every product receives a continuously updated robustness score derived from process capability indices, material quality signals, deviation trends, stability data, and manufacturing variability. Risk scoring agents aggregate signals across the quality ontology to identify products approaching quality boundaries before they cross them — enabling proactive mitigation that prevents write-offs, supply disruptions, and regulatory findings.

  • Continuous Cpk/Ppk trending and multivariate statistical analysis across all critical quality attributes for every product
  • Predictive risk models correlating material variability, process drift, and quality outcomes to forecast failures
  • Product robustness scoring that quantifies quality margin across the entire portfolio with drill-down by CQA
MES

Standardised Quality Data Collection Across Every Manufacturing Site

Deploy consistent batch record templates, in-process control workflows, and quality data collection standards across all manufacturing sites. When every facility captures data in the same format, at the same granularity, and against the same specifications, cross-site quality comparison becomes meaningful and the integration pipeline operates without site-specific data transformations.

  • Standardised electronic batch records with built-in quality data capture at every production phase
  • Consistent in-process control workflows and specification templates across all facilities
  • Real-time production data streaming into the integration pipeline as batches progress
FDA Tracker

Regulatory Intelligence for Quality Assessment Expectations

Monitor how FDA cites product quality assessment failures across the industry. Understand which process validation gaps, statistical deficiencies, and risk management shortcomings draw 483 observations and warning letters — so your team can proactively align practices with current inspector expectations before the next audit.

  • Real-time tracking of 483 observations related to process validation, CPV, and quality assessment deficiencies
  • Warning letter analysis for product quality, risk management, and statistical control shortcomings
  • Benchmarking your quality assessment practices against industry enforcement trends and peer performance

Results

Measurable Impact

Real results from organisations using Celestara for continuous product quality assessment and predictive risk management

10x
Risk Detection Speed
Faster identification of emerging quality risks through continuous monitoring versus annual retrospective CPV reviews
95%
Data Consolidation
Quality-relevant data sources unified into a single ontology-backed model, up from typically 30–40% coverage in manual processes
12%
Right First Time
Absolute improvement in right-first-time rates through predictive risk mitigation and proactive process adjustments
40%
Write-Off Reduction
Decrease in batch write-offs and reject events through early risk signal detection and material-quality correlation

Next Step

Get Started

Stop discovering quality risks after the write-off. Celestara AI agents continuously assess product robustness across your entire portfolio — correlating process data, material attributes, and quality events through a unified quality ontology to predict risks before they materialise. Shift from reactive compliance to predictive quality intelligence.

Get Started
Exit