PHARMA AI WORKSPACE

PHARMA AI WORKSPACE

Local-First AI for Indian Pharma Operations

Free • Private • Runs on Your Machine

The Problem We Solve

Indian pharma companies operate in one of the world's most regulation-dense environments — CDSCO approvals, UCPMP compliance, government tenders, multi-language field communications, and mountains of SOPs — yet the AI tools available today are built for generic business users, not pharma professionals.

The result: critical tasks that should take minutes take hours. Compliance mistakes slip through. Institutional knowledge sits locked in PDFs nobody can query. Doctor call insights never make it into a CRM. And sensitive molecule data gets pasted into cloud tools that were never designed for pharmaceutical confidentiality.

Pharma AI Workspace was built to fix exactly that.

What It Is

Pharma AI Workspace is a desktop AI productivity suite with 11 purpose-built modules that cover the full pharma operations workflow — from product naming and regulatory reading to doctor call transcription and tender generation. It installs with a single .exe file. It runs entirely on your machine. It costs nothing to operate.

It is not a generic chatbot with a pharma label. Every module is built around the specific regulations, formats, and language of the Indian pharmaceutical industry — CDSCO naming conventions, GeM tender formats, UCPMP compliance rules, Indian Contract Act clauses, and 12 Indian languages are all built in by design, not bolted on as an afterthought.

11 Modules. One Workspace.

Module

What It Does

Product Naming

CDSCO-compliant brand name generation for new molecules

Multi-Language Translator

Pharma-accurate translation across 12 Indian & international languages

Agreement Drafter

NDA, distribution, CMO & licensing agreements under Indian law

Regulatory Reader

Instant extraction of action items from CDSCO / FDA / EMA circulars

Meeting Minutes

Live recording or audio upload → structured, AI-formatted minutes

SOP Q&A Bot

Ask questions directly against your uploaded SOP documents

Visual Aid Analyser

UCPMP compliance scoring on promotional and marketing materials

Call Summaries

Doctor call transcription → CRM-ready structured summary

Opportunity Scanner

openFDA + PubMed queries → instant market opportunity brief

Tender Generator

Government tender response documents in GeM & state formats

WhatsApp Personaliser

HCP message personalisation at scale with UCPMP scoring

Built on a Privacy-First Architecture

In pharma, data confidentiality is not a feature — it is a requirement. Molecule pipelines, doctor relationships, internal SOPs, and pricing strategies are among the most sensitive assets a pharma company holds. Pharma AI Workspace was architected with this reality at its core.

• Audio recordings of doctor calls never leave your PC — transcription runs locally via Whisper.cpp, a compiled C++ binary bundled inside the app

• API keys are encrypted using your operating system's native keychain (Windows DPAPI), not stored in plain text

• A strict Content Security Policy blocks all outbound connections except to the exact APIs the app uses — no telemetry, no analytics, no background data collection

• No account required. No login. No data stored on any server we control.

Intelligence That Adapts to Your Setup

Not every pharma office has the same infrastructure. Pharma AI Workspace uses a three-layer AI dispatch system that automatically routes to the best available backend — no configuration required.

• If you have a local Ollama model installed, it runs there — fully offline, zero API cost

• If local is unavailable, it falls through to OpenRouter's free tier automatically

• If OpenRouter is rate-limited, it falls through to Google Gemini's free tier

For large-document tasks — SOP analysis, regulatory circulars, PDF contracts — the system intelligently prioritises Gemini's 1-million-token context window, so even 100-page documents are processed in a single pass. The app never crashes because one API is down. It always finds a path.

Compliance Is Non-Negotiable — So We Made It Automatic

UCPMP violations don't just create legal risk — they damage relationships with regulators and healthcare professionals built over years. Every piece of promotional content that leaves your field force should be checked. In practice, it rarely is.

Pharma AI Workspace embeds a deterministic UCPMP compliance engine that runs on every message and visual aid before it is output:

• Forbidden superlatives are blocked — best, safest, most effective, number one, cure, miracle

• Unsubstantiated percentage claims are flagged automatically

• Comparative claims trigger a data-on-file warning before the content is used

• Every output receives a compliance score from 0 to 10 with specific issues listed

This is not AI being asked to 'check for compliance.' It is a rule engine that cannot be confused, convinced, or hallucinated away from the facts.

MAI — Your Pharma Operations Assistant

At the centre of the workspace is MAI, a conversational AI assistant that understands pharma operations. Ask MAI to name a molecule, draft a distribution agreement, generate WhatsApp messages for a cardiologist, or scan an opportunity — it gathers what it needs through natural conversation and executes the right module automatically.

Users do not need to know which module to open. They talk to MAI and MAI does the work.

Zero Operational Cost. Forever.

The free tiers of OpenRouter and Google Gemini together provide enough capacity for an entire pharma ops department to run every module, every working day, at no cost. Audio transcription through local Whisper is unlimited with no API involved at all. FDA and PubMed queries are free public APIs with generous rate limits.

Pharma AI Workspace was designed so that the answer to 'what does this cost to run?' is always: nothing.

Ready to Deploy in 10 Minutes

• Download the installer — one .exe file, no dependencies

• Follow the first-run wizard — Whisper model downloads automatically in the background

• Enter a free OpenRouter key and a free Google Gemini key

• All 11 modules are immediately available — no training, no onboarding sessions required

• Future updates install silently in the background — no IT involvement needed

Pharma AI Workspace is not another AI tool for pharma.

It is the first AI workspace built from the ground up for the way Indian pharma actually works.

AI-Powered Social Media Automation for Filipino Community Creators

The Creator's Problem

Building a consistent, high-quality Instagram presence for a Filipino community audience is a full-time job. Scripts need to be written in Taglish. Videos need to be assembled. Captions need hashtags that actually work. News needs to be found, verified, and turned into content before it goes cold. And none of it can sound like it came from a generic AI tool — because Filipino audiences immediately recognize and reject content that doesn't speak their language.

Palawan Creator automates the entire pipeline. From finding the news to posting the Reel, every step is handled automatically — on a schedule, in the right voice, in the right format, at the right time of day.

Seven Pillars. Seven Voices. One System.

Palawan Creator operates across seven content pillars — Local News, Job Alerts, OFW Support, Palawan Cuisine, Nature & Science, Environmental Conservation, and Government Programs. Each pillar has its own AI persona with a distinct Taglish voice: Manong Ronnie for news, Ate Cita for jobs, Nanay Mely for recipes, Kuya Jun for OFW families, and more.

Audiences don't experience a brand posting at them. They experience a neighbor talking to them. That is an intentional design decision — and it is why this content performs.

What the System Does — Automatically

• Monitors six RSS feeds and scores articles by local relevance, engagement potential, and content safety — no manual topic selection

• Writes scripts in the voice of the assigned pillar persona, then passes every script through a five-gate constitutional AI safety check before it is used

• Generates captions with exactly 25 Taglish hashtags within Instagram's character limit

• Converts scripts to audio using Microsoft's Filipino neural voice (BlessicaNeural) with word-level timestamps for frame-accurate subtitles

• Assembles portrait 1080x1920 MP4 Reels with stock footage, voice track, and burned-in subtitles via bundled FFmpeg

• Posts to Instagram via the official Graph API on a PHT-timezone schedule — with automatic retry for any failed posts within a 4-hour recovery window

• Detects breaking news spikes and triggers the relevant pillar outside the normal schedule

• Accepts voice commands in Taglish — so the creator can control the pipeline from anywhere, without touching a screen

Built on a Foundation That Never Goes Down

Every critical system in Palawan Creator has a fallback. The LLM dispatcher chains across Gemini, Groq, OpenRouter, and Pollinations automatically — if one provider rate-limits or fails, the next takes over without interrupting the pipeline. The scheduler detects and recovers missed posts. The topic pool is pre-generated so posting never waits on real-time scoring. The system is designed to keep running even when individual components do not.

Privacy and Security by Design

All API credentials are stored encrypted using Fernet symmetric encryption, tied to the user's machine. No credentials are stored in plain text anywhere in the application. Changing the account's content niche requires a PIN to prevent accidental misconfiguration. Each account's data — topics, post history, settings — is stored in isolated user data directories that survive app reinstalls.

Zero Cost to Operate

Palawan Creator runs entirely on free-tier services — Gemini Flash, Groq, Edge TTS, Pexels, Pollinations, and the Instagram Graph API. A creator posting multiple Reels per day across seven content pillars pays nothing in API fees. The only infrastructure required is the Windows machine the app runs on.

One Installer. Ready in Minutes.

Palawan Creator ships as a single Windows executable. Python, FFmpeg, and all dependencies are bundled inside. The first-run wizard walks through API key setup and Instagram authentication. From installation to first automated post takes less than 20 minutes. Updates install silently in the background — no creator action required.

Palawan Creator gives a solo community creator

the content output of an entire production team —

without the team, without the budget, and without the hours.

MAI

M.AI 0.1

Your AI. Your Infrastructure. Your Rules.

Voice + Text Assistant • 8 Providers • Persistent Memory • GitHub Awareness • Zero Cost

Why It Matters

Every mainstream AI assistant comes with the same trade: convenience in exchange for control. Your conversations live on someone else's servers. Your API access is metered by someone else's billing. Your assistant forgets you the moment a session ends — unless you pay for it to remember.

M.AI 0.1 rejects that trade entirely. It is a personal AI assistant that runs on infrastructure you control, orchestrates the free tiers of eight LLM providers into one seamless intelligence, remembers everything across sessions, and operates under a written constitution — at a running cost of exactly zero.

One Assistant. Eight Brains.

Behind M.AI's single chat window sits a capability-aware routing engine connected to Google Gemini, Groq, Cerebras, SambaNova, OpenRouter, Hugging Face and more — over twenty models in total. Every request is routed to the provider best suited for the task: lightning-fast models for voice replies, deep reasoning models for hard problems, dedicated code models for programming.

When a provider fails or rate-limits, the router silently moves to the next. A self-healing health system probes degraded providers every five minutes and restores them automatically. The user never sees an outage — they just see answers.

It Knows the Whole Portfolio

M.AI is connected to its creator's GitHub account — every repository, every project, every README. Ask it what any project does, how it's built, what technology it uses, or how two projects compare, and it answers from the actual source code, not from a stale description. It is a living, conversational index of an entire body of work — and for anyone visiting, it means the fastest way to understand any project is simply to ask.

It Remembers You

M.AI runs a three-tier memory system. Session context keeps the current conversation sharp. Project knowledge gives it deep awareness of the work you've connected. Long-term memory — stored with vector embeddings and recalled by semantic similarity — means it remembers conversations from weeks ago and brings them back exactly when relevant. Memory persists across restarts, on your storage, owned by you.

It Does Real Work

M.AI is not limited to talking. Connected projects expose skills — start a dev server, run a build, trigger a content pipeline — that the assistant executes in sandboxed processes with strict CPU, memory, and timeout caps. Its first integrated project is a complete Instagram automation system that it can drive end-to-end by voice command.

Speak to it or type to it: dual speech-to-text paths and browser-native voice output mean the entire voice experience works without a single paid speech API.

Security Without Trust

M.AI uses a bring-your-own-keys model with a guarantee most platforms cannot make: your API keys never touch the server's storage. Keys live in your browser, travel as per-request headers, are used for one call, and are discarded. No accounts, no key vaults, no server-side secrets. Even on a hosted instance, the operator never holds your credentials.

Governed by a Constitution

Nineteen written rules are compiled into the assistant's core at startup: amplify the human, never replace them. Log everything. A panic stop command is always valid and immediate. Never use a paid service. Self-modification requires a six-stage review gate with human approval. The assistant's boundaries are not implied — they are documented, versioned, and enforced.

Deployed Like Software Should Be

One Docker Compose file runs the entire system — orchestrator, web UI, and Redis — identically on a laptop, a budget VPS, or Oracle's free ARM cloud. It is live in production today, served over SSL, built for both ARM and x86. Setup is a clone, an install, and a single command.

M.AI 0.1 proves a simple thesis:

a personal AI you fully own can be as capable as one you rent —

and it can cost nothing at all.

JOBHUNT AGENT

The Self-Hosted AI Job Application Automation Platform

Automate Your Job Search • AI Resume Tailoring • 100% Private • 100% Free

Automate Your Job Search Without Giving Up Your Data

Searching for a job is a full-time job. Sourcing openings across dozens of career pages, tailoring a resume for every application, filling the same forms again and again, tracking follow-ups — the repetitive work consumes the hours that should go into interview preparation.

JobHunt Agent is a self-hosted job application automation tool that handles the entire workflow — job sourcing, AI-powered job matching, resume tailoring, automated form filling, and application tracking — while keeping every piece of your personal data on your own computer. No cloud uploads. No subscriptions. No third party ever holds your resume.

AI Job Matching That Explains Itself

JobHunt Agent automatically sources job listings from official Greenhouse, Lever, Ashby, and Workable APIs plus remote job boards like Remotive and Arbeitnow. Every job is scored by AI against your skills, target roles, and preferences — with a written reason, a list of matched skills, and a list of missing skills for every score. Duplicate listings across job boards are detected automatically using vector embeddings, so each opportunity appears exactly once.

AI Resume Tailoring That Never Leaves Your Machine

For every shortlisted role, JobHunt Agent generates a tailored resume and a custom cover letter using a local AI model running on your own hardware. Your master resume, your work history, and your personal details are never sent to any cloud service — the privacy boundary is built into the architecture itself. Cloud AI sees only the public job posting text. You see professionally tailored application documents for every role.

You Approve Every Application — Always

JobHunt Agent is not a spam cannon. Every prepared application enters a review queue where you see the job, the score, the tailored resume, and the cover letter — and nothing is submitted until you explicitly approve it. Once approved, browser automation fills the application form, attaches your documents, and captures a screenshot for your records. A built-in idempotency system makes it technically impossible to apply to the same job twice.

Application Tracking and Follow-Up, Handled

Every submitted application is tracked automatically. Follow-up reminders are scheduled seven days out. Stuck applications are flagged. A daily digest lands in your Telegram with everything that happened: new jobs sourced, scores assigned, applications submitted, and items waiting for your review. Thirteen generated PDF reports — from a top-scoring shortlist to a skills-gap analysis that tells you exactly which skills to learn next — turn your job search into measurable progress.

Private by Architecture, Free by Design

The entire platform runs on your machine in Docker containers bound to localhost — nothing is exposed to the internet. Credentials are encrypted at rest. Logs automatically redact personal information. And the entire stack is built from free and open-source components: PostgreSQL, Ollama local AI, Playwright browser automation, and self-hosted n8n orchestration. There is no subscription, no per-application fee, and no premium tier. The operating cost is zero.

Set Up in Minutes, Runs on Schedule

One docker-compose command brings up the full 11-service stack. Add your master resume, set your target roles and skills in the dashboard, and the pipeline runs automatically on a schedule — sourcing, scoring, and preparing applications around the clock, then waiting for your approval before anything goes out. Your job search keeps moving even while you sleep. The decisions stay yours.

JobHunt Agent does the hundred hours of repetitive work

so you can spend your energy on the one hour that matters — the interview.

EXAMINER AI

AI Exam Grading for Descriptive Answers — Explainable, Rubric-Based, Private

Automated Answer Evaluation • Instant Student Feedback • Local AI • Zero API Cost

Automated Exam Grading That Shows Its Work

Grading descriptive exam answers is the slowest, most expensive, and most inconsistent process in education. A trained examiner spends minutes on every long-form answer; a board exam generates millions of them. Students wait weeks for results and receive a number with no explanation. And two examiners grading the same answer routinely award different marks.

Examiner AI is an AI-powered exam evaluation system that grades descriptive and numerical answers in seconds — using the same structure a human examiner uses: a marking rubric, partial credit, step verification, and penalty deductions. Every score comes with a transparent, criterion-by-criterion breakdown showing exactly where each mark was earned or lost.

Three Signals. One Defensible Score.

Most AI grading tools ask a chatbot for a number and hope. Examiner AI triangulates three independent evaluation signals: a deterministic rubric engine aligned to Maharashtra SSC marking standards, neural semantic similarity that measures whether the student understood the concept — even in their own words — and AI-generated feedback that explains the result. The rubric carries the most weight; the AI never grades alone.

A student who paraphrases a correct answer still earns marks, because the system compares meaning, not vocabulary. A student who skips a solution step loses exactly the marks that step was worth — and is told so.

Instant Feedback Students Can Learn From

Every evaluated answer receives constructive, educator-style feedback generated by a locally-running AI model: what was done well, which key points are missing, and what to improve next time. Grading stops being a verdict and becomes a teaching moment — delivered in seconds instead of weeks.

Handles the Answers That Matter

• Long-form theory answers — solution structure, diagram presence and labeling, language clarity

• Short definition answers — keyword and concept coverage with penalty deductions

• Numerical problems — formula correctness, substitution steps, unit validation, and final-answer tolerance within ±2%

The rubric engine is fully extensible — new answer types and marking schemes plug into the same scoring pipeline.

Private by Architecture

Student exam data is among the most sensitive data an institution holds. Examiner AI runs every AI component locally — the embedding model and the feedback LLM both execute on the institution's own hardware through Ollama. No student answer, no question paper, and no grade ever touches a cloud API. There are no per-evaluation fees, no API keys, and no external dependencies in the grading path.

Built for Trust, Not Just Speed

Every evaluation carries a confidence score, and the system is designed for human-in-the-loop review: low-confidence cases escalate to a human examiner rather than being silently auto-graded. Every rubric result exports as a structured audit record — so when a grade is questioned, the institution can show precisely how it was computed. Examiner AI is built to assist examiners and earn the trust of students, parents, and boards.

Weeks of grading, delivered in seconds —

with every mark explained, every answer private, and every educator still in control.

PLANT DISEASE DETECTION

AI Crop Disease Diagnosis With Treatment and Dosage — From a Single Leaf Photo

38 Diseases • 14 Crops • Instant Diagnosis • Exact Dosage • Weather-Aware Advice

AI Plant Disease Identification From One Photo

Crop diseases destroy up to 40 percent of global agricultural production every year — and for most farmers, getting a reliable diagnosis means waiting days for an expert, guessing from a search engine, or spraying the wrong chemical and hoping. By the time the disease is identified, it has often already spread.

Plant Disease Detection is an AI-powered crop disease diagnosis system that identifies 38 plant diseases across 14 crops — tomato, potato, corn, grape, apple, and more — from a single leaf photograph, in seconds. The deep learning model, built on ResNet50 transfer learning and trained with field-realistic image augmentation, recognizes fungal, bacterial, and viral diseases as well as pest damage, even in imperfect phone photos.

From Diagnosis to Treatment Plan — Automatically

A disease name alone does not save a crop. For every diagnosis, the system delivers a complete, actionable treatment plan: the specific fungicide or pesticide product for that disease, step-by-step application instructions, spray schedules, prevention strategies for the next season, and expert tips on timing and safety. Disease identification becomes disease management.

Exact Dosage for Your Exact Field

Pesticide overdosing wastes money and damages soil; underdosing lets the disease survive. Enter your field size in hectares, and the system automatically calculates the precise quantity of product needed — based on the recommended concentration and standard water requirements per hectare. The built-in cart turns that calculation into a ready shopping list. No guesswork, no waste, no under-treatment.

Smart Farming Advice That Knows the Weather

The advisory layer integrates a 7-day weather forecast — temperature, humidity, and rainfall from NASA satellite data — directly into the recommendation. A locally-running AI agronomist combines the diagnosis, the forecast, and the field context into practical guidance: spray before tomorrow's rain, adjust for crop maturity, monitor for secondary infections. Optional satellite vegetation indices add field-level crop health monitoring. This is precision agriculture made accessible.

Serious Machine Learning Underneath

The model is built like production ML, not a demo: ResNet50 transfer learning with a deep custom classification head, two-phase fine-tuning that protects pre-trained knowledge, aggressive augmentation simulating real field photography, mixed-precision GPU training, and rigorous per-class evaluation with full confusion-matrix analysis across all 38 classes. A lightweight 25 MB inference model makes deployment affordable — from cloud hosting down to edge devices.

Built for Real Farms

Every design decision targets practical farming: treatment recommendations drawn from products farmers can actually buy, dosage mathematics in hectares and liters, weather data from a free public API, and AI advice that runs on local hardware with zero per-query cost. The entire system operates without subscription fees — the technology serves the farm, not the other way around.

One photo. One diagnosis. One complete plan —

delivered in the seconds between noticing a sick leaf and deciding what to do about it.