Hardik Joshi

Compliance Risks in Pharma Formulation R&D
Even though well-meaning messages in emails, reports, or submissions to regulators, these could break compliance rules in drug formulation research. An AI helper could stop:
Regulatory Mistakes: Risk: Making inappropriate statements about how well a drug works?, how safe it is?, or how it functions? (breaking FDA/EMA rules). Example: Statement: “This mix cures Disease X”. As per correct sense, a statement should be “The drug shows promising lab results against Disease X.” Leaks of Secret Information: Risk: Sharing private details about how a product is made (like exact ingredient amounts, new delivery methods) with people who shouldn't know. Example: Statement: “We use 12% Polymer Y to control release”.  As per correct sense, a statement should be “a polymer-based system to control release.” Patient Info & Privacy (HIPAA/GDPR): Risk: Talking about details of people in clinical trials (even hidden info can be risky if not removed). Example: Statement: “Patient 45, a 60-year-old man, did well”. As per correct sense, a statement should be “Some people in Group B got better.” Wrong Paperwork: Risk: Not following ICH/GLP rules (like not reporting all stability data, missing records of batches). Example: Statement: “Early data hints at stability”.  As per correct sense, a statement should be “Mid-term stability data (40°C/75% RH, 3 months) shows no big breakdown.”. Contract & Partner Dangers: Risk: Promising too much to partners (like “We'll hand in the NDA by Q2” before checking) How AI Can Give Gentle, Unobtrusive Input:
To keep scientific work flowing, AI should act as a “quiet compliance helper”:
Smart Highlighting in Drafts: In-text Alerts: Underline risky words with color coding: Red: “Rule warning: 'cure' hints at unapproved health claim.” Yellow: “IP heads-up: Only list excipient if your readers need to know.” Mouse-over Tips: “shows lab-based antiviral action instead of 'kills Virus Y.” Compliance-Friendly Word Completion: When you type words like “safe” or “effective” AI offers pre-approved wording: "Safe": “Showed good tolerance in early tests” "Proprietary": “A new fat-based delivery method” (keeps formula secret). Quick Fixes for Common Mistakes: A “Fix It” button (in email or lab software) rewrites flagged parts with little effort. Quiet Alerts for High-Risk Text: If AI spots big issues (like missing side effects in a draft report), it tells the quality team, without stopping the writer. On-the-Spot Learning: A “Why This Matters” link in warnings takes users to company rules or drug guidelines (like “Q6A Specs: Why we don't say 'pure' in descriptions”).

AI-Driven Cross-functional Team Pattern Recognition to Gain Company Insights
The Problem
Bosses get scattered reports on issues, problems, and findings from HR, Finance, Operations, and Tech Support. Each team keeps track of its own numbers, but big inefficiencies or repeat issues might span several departments yet stay hidden in separate reports.
AI Fix: A Prompt + Flow-Based Pattern Finder
An improvised AI infrastructure could take in organized data from all departments, use NLP and grouping methods to find hidden links.
How It Works
(1)   Input:
·        Numeric data:
HR: People leaving, days off, slow promotions Finance: Surprise expense, slow bill approvals Operations: Delayed timeline, not enough stock Tech: Same problems persist, failed updates ·        Non-numeric data:
How workers feel (checking Slack/MS Teams talk) Support tickets from users ("VPN crashes all the time!") Incident reports ("Finance put a hold on funds mid-project") ·        Background information:
Time stamps (to spot recurring patterns) Groups/jobs involved (to trace who's responsible for what) (2)   AI Processing: Joining the Pieces
·        Phase 1: Signal Collection:
Standardizes information (e.g., "slow system" in HR exit interviews → connects to Tech's "tickets"). ·        Phase 2: Trend Identification
Time-Based Grouping: "Each quarter, Finance holds up → Ops hurry → Tech make mistakes." Cause-Effect Links: "People quit more often 3 weeks after big system failures." Unusual Event Spotting: "Why is this month's Ops delay 4 times higher when other metrics stay the same?" ·        Phase 3: Idea Creation
AI proposes root causes (e.g., "65% of HR complaints talk about 'approval tied to Finance's new process"). (3)   Results: Insights Bosses Can Use
·        Live "Company Health" Screen
Pressure Maps: Show friction between departments (e.g., "When Finance turns red, HR becomes yellow 2 weeks later"). Future Warnings: "Tech's backlog will reach dangerous levels in 10 days unless Ops stop new requests." Response Plans: "For 'Money freeze → People leaving' patterns, tell everyone about roadmap changes ." ·        Weekly "Patterns to Watch" Brief:
Top 3 Recurring Themes: "VPN slowness → Loss of productivity → HR complaints" "Quarter-end Finance crunch → Ops shortcuts → Tech debt" One Strategic Fix: "Line up Tech/Finance roadmaps to stop Q4 from happening again." Example Scenario
AI notices that whenever Tech rolls out a big update without talking to Ops/Finance:
Ops fails to meet deadlines (because of surprise work), Finance goes over budget (rush approvals), HR gets feedback from leaving staff about "messy priorities." Advice: Make sure teams check how updates affect others before launching.*
What Sets It Apart
More Than Just Numbers: It digs deeper than charts to explain why patterns show up. It Tells a Story: It breaks down insights in plain language, not just data. It Keeps Getting Smarter: It fine-tunes its advice as it sees results. This changes scattered info into a smart early-warning setup—where AI doesn't just flag issues, it helps stop them before they start.

Prompt-Flow Systems for AI-Driven Excipient Compatibility Screening
What Makes This Unique:
In pharmaceutical formulation R&D, choosing appropriate excipients (inactive components that aid in the delivery of pharmaceutically active substances) is often crucial and very tedious. Making the wrong selection may lead to problems with stability, efficacy, or complete failure of clinical trials. This step has historically depended on thorough manual literature reviews, trial-and-error experimentation, and expert deep intuition.
This task can be performed by an AI system using prompts and flows to automate and optimize.
Exemplar interactions:
·       Structured prompts support predicting interactions.
·       LLM knowledge retrieval support enabled predictive formulation hypothesis generation.
·       Guided lab workflows enable dynamic decision logic algorithms for guiding workflows.
Description: How It Works - A prompt-flow orchestration 
1. Knowledge Retrieval & Hypothesis Generation (Prompt Layer)
Input Prompt:
"Available [Drug Molecule: X] with specific properties like low pH sensitivity and poor solubility, identify constituents compatible with respect to FDA’s formulated excipients records, avoiding known incompatibilities and encounter discriminative stability enhancement order. Rank by Stability Enhancement Potential.” 
AI Action: 
AI sends queries to Internal Databases and regulatory agencies sources. 
Literature (RAG – Retrieval-Augmented Generation) and cross-reference outputs list as “Mannitol (high compatibility), PVP (medium), exclude other PEG and stearic acid).”
2. Dynamic Experiment Design (Flow Logic)
Decision Node: If the AI proposes a novel excipient,
·       QSAR models will trigger stability simulation. For example, the use of an uncommon polymer would fall under this category.
·       Revises approval precedence (EMA/FDA approval history).
·       If confidence is high, sends to lab automation systems for bench testing.
3. Revised Iterative Optimization Loop
Feedback request: "Would suggest alternatives or modifiers for prior formulation with [Excipient Y] cited [Issue: precipitation at pH 5]."
Every new experiment and data point modifies the next recommendation the system gives.
Why This is High-Impact:
·       Excursion screening significantly reduces formulation development time.
·       Prevention of significant errors is achieved through flagging incompatibilities before lab trials.
·       Inexperienced researchers become recipients of AI-system recommendations by achieving cost democracy.
·       Compliance checks were built on which ensure the confidence of regulations

AI Agent for Pharma R&D Literature Search
Key capabilities:
1. Raw Contextual Grasp:
a. Analyzes scientific literature such as papers and clinical trials at a near-human level and derives hypotheses, methodologies, and results even when relying on incomplete or self-contradictory information.
b. Point out implications without a clear or logical basis towards hypothesis suggesting (“X off-target effect by Compound X in The Cell paper is consistent with unpublished toxicity data”).
 
2. Hypothesis Generation:
a. While integrating (cross-referencing) unrelated fields, propose new theory directions. Ex: Alzheimer’s mechanism in Nature could apply to your Parkinson’s work, here’s a synthesis pathway.
b. Anticipate no innovation “white space” opportunities. For example, Cyst inhibition has never been investigated for this rare cancer subtype.
 
3. Real-Time Collaborative Curation:
a. Act as a co-scientist alongside researchers as a thought partner who participates to dynamically update during meetings (new Preprint just dropped that challenges your target—do you want to review?).
b. Create visual summaries, including but not limited to competing drug mechanisms such as interactive graphs.
 
4. Regulatory and Competitive Intelligence:
a. Show awareness for worldwide rule setting, such as FDA’s new guidance on digital endpoints has an impact within phase III design, and how it relates to the supervision of competitor pipelines, like rival Y, has dropped this target because they showed signals of unsafe markers.
 
5. Oversight Self-Validating Citations:
a. Citations would need evaluation based on rationale and scores given to documents where they achieved self-revalidation of the set hypothesis, indicating that they exceeded verification of their arguments.
 
AI interaction with humans:
1. For Researchers:
a. Voice/chat interface ("List me all active patents related to molecule A that have a controlled release drug profile”).
b. "Pop-up" notifications for important updates ("Reference product is delisted from the US market due to potential adverse effect").
 
2. For Executives:
a. Generates a clear report stating risk and benefits involved with the diagram ("Here’s why launching product A aligns with the current product portfolio in the market.").
 
Potential Risk to Guard Against:
Overlooking important information:
· The AI can be overdependent on "highly reviewed articles from reputed papers" or institutional biases (e.g., ignoring new research or recent findings due to low or no popularity).
Risk mitigation plan:
· Add special prompts that force AI to search the entire information irrespective of popularity, review, or rating.
· Mandatory requirement of human signature or confirmation, especially in high-risk recommendations. (e.g., clinical trial design changes)

Scenario: Error in AI-driven R&D Formulation Development
Scenario: A Pharma R&D organization uses an AI system to generate formulations for a new Anti-cancer medication. The AI suggests a formulation with an upper limit of excipient, provided by the USFDA, which claims that it improves efficacy and stability. The formulation clears automatic system checks but is later found to cause mild adverse reactions (e.g., vomiting and acidity) in a few volunteers during trials. The investigation confirms that the AI was trained mostly on data from healthy young populations, but it was missing sensitivity in paediatric and geriatric populations with different metabolisms.
This is a critical matter due to the involvement of patient safety, regulatory risk, and reputation damage.
 
Assigning Responsibility
(1)   AI Engineer – Prime responsible - Due to not including diverse population data e.g., paediatric and geriatric patients. Also, not setting stringent standards for the safety of excipients.
(2)   R&D Scientists – Scientists have conducted trials without validating facts regarding excipients.
(3)   Complinace Teams – The Team has skipped review due to confidence on AI system.
 
Design Safeguards for Transparency & Traceability
Training Data Data should include all patient demographics (age, gender, comorbidities) and include treatment-specific data for edge cases (e.g., patients with gastric disorders). Confidence score for Formulation Decisions The AI should output not just a recommended formulation, but also include confidence scores (e.g., "90% as highly recommended, but only 70% recommended for elderly patients"). Human interference for Critical Decisions Suggested formulation is near safety thresholds (e.g., excipient limits) and must require the scientist's sign-off. Real-time Feedback Integration Model also updates data simultaneously, including all adverse events. This approach helps AI-driven formulation development without compromising safety, keeping accountability on the humans who design, develop, validate, and use the system.
 

Status quo bias is defined as the continuing current strategy or idea and rejecting new strategy or ideas even though it may change state of affairs. 
It was first identified by Mr. William Samuelson and Mr. Richard Zeckhauser in 1988 in the academic article “Status Quo Bias in Decision-Making.” They have done various experiments regarding given the choice between the status quo and a new option and they found that people were more likely to follow already existing situation.
It is the situation where decision is taken emotionally rather than logically. Whenever changes occur or proposed then people may feel uncomfortable where the outcome is unknown. This tendency is to keep people themselves safe and the things continue as it is. 
Ultimately status quo bias negatively affects ability to make decisions. Due to this people may miss valuable opportunities. Status quo bias can also hinder the growth of a business. If top management is not willing to take risks which could benefit the business at large, the company could begin to stagnate. 
Examples of Status Quo Bias in the Workplace?
If a company already have CDA with vendor and CDA is about to expire. Then company will renew agreement without giving opportunity to another vendor which may be better than earlier. 
When a company is introducing SAP system to replace ERP then employee show preference for keeping existing ERP for which they are more familiar. However, in actual scenario, SAP provides better flexibility and more options for management.
How to Overcome Status Quo Bias?
If you’re in a leadership position then it is up to you that how to communicate organizational change to your team. 
Learn to recognize status quo bias in yourself and others
It is a leadership skill that understand status quo bias when it happens.
Weigh the advantages and disadvantages
Sometime status quo bias may be beneficial to take wrong decision. So, we need to identify pros and cons for this. Collective decision needs to be taken. 
Frame the default option as a loss 
We need to consider loss as a default option according to loss aversion principle.
Follow the REDUCE framework 
Leader need to establish certain framework that reduces barrier for change.