For faculty, doctors, trainees, and students.
This page is a collection of AI tools and resources for faculty, doctors, trainees, and students. Tools change quickly; treat this as a starting map, not an endorsement.
The keynote covered the brief history of AI, how large models actually work, major breakthroughs, ethical risks (including hallucinations and privacy), and what this means for the clinician of tomorrow.
Open presentation in full window
Pick the section that matches your job for today. Each table lists several commonly used tools and links to access them. Prefer tools with a free or education tier when you are exploring. Never paste identifiable patient data, unpublished hospital records, or passwords into public chat tools.
Start here if you are new to these tools. Conversational models are useful for drafting and explaining; research assistants are better when you need papers; notebook-style tools are better when you want answers grounded in files you already have.
A reliable medical prompt usually states: (1) your role, (2) the audience, (3) the task, (4) constraints (guidelines, word count, language), (5) “cite sources or say you cannot”, and (6) “state uncertainty and what a human must check”. Example: “You are a faculty member writing for third-year MBBS students in Bangladesh. Draft 8 OSCE-style questions on neonatal sepsis. Flag any item that needs a local guideline check. Do not invent citations.”
| Task / use case | What to use AI for | Tools (commonly used) | Tips & caveats |
|---|---|---|---|
| General drafting and explanation | Outlines, first drafts, simplifying a concept, translating, generating draft quiz stems. | ChatGPT, Claude, Gemini, Microsoft Copilot, Grok | Free tiers exist but rate-limit. Paid plans add longer memory and file upload. Always verify clinical facts. |
| Web-grounded answers | Quick orientation to a topic with links you can open and read yourself. | Perplexity, Gemini (with Search), ChatGPT (with web search) | Citations can still be wrong. Open the source; do not cite the chatbot. |
| Work from your own PDFs and notes | Summarise lecture notes, guidelines you uploaded, or a folder of papers you already collected. | NotebookLM, Claude (Projects / files), ChatGPT (custom GPTs / files) | Upload only materials you are allowed to share. NotebookLM is strong for audio overviews of your sources. |
| Paper-first literature tools | Find related papers, extract methods, map a field before you start a review. | Elicit, Consensus, Scite, Semantic Scholar, Google Scholar Labs, PubMed | Use these to find papers, then read the papers. They do not replace a systematic search protocol. |
| Open or lower-cost models | Experiment when paid APIs are unaffordable, or when you want a locally hosted option later. | DeepSeek, Mistral, Qwen, Ollama (local) | Local tools need a capable computer. Privacy is better on-device, but quality and medical safety still need human review. |
Faculty and tutors can use AI to draft materials faster, then spend time on accuracy, pedagogy, and local context (curriculum, language, available investigations). Students can use the same tools to study (not to outsource learning).
| Task / use case | What to use AI for | Tools (commonly used) | Tips & caveats |
|---|---|---|---|
| Classroom lectures and PowerPoint / Google Slides | Turn a topic outline into slide titles, speaker notes, learning objectives. |
General:
ChatGPT,
Claude,
Gemini.
Slides: Gamma, Beautiful.ai, Slidesgo, PowerPoint Copilot, Google Slides with Gemini |
Auto-generated slides often look finished but miss local epidemiology, drug names, and your exam blueprint. Keep one “facts to verify” slide for yourself. |
| Infographics and visual explainers | Draft a figure brief, then generate diagrams for pathophysiology, pathways, or patient education. | Canva Magic Studio, Figma (AI features), Napkin.ai, BioRender (scientific figures), Gemini image generation | Anatomy and drug structures are frequently wrong in generated images. Label every figure as “illustrative” until a clinician redraws or checks it. |
| Short videos and animations | Script a 2–3 minute explainer, storyboard scenes, generate a voiceover, or assemble a simple screen-capture lesson. |
Scripting: ChatGPT / Claude / Gemini.
Video: Descript, CapCut, Clipchamp, Synthesia (avatar, paid). Voice: ElevenLabs |
Avatar tools are costly and can feel impersonal for clinical teaching. A recording of you speaking plus captions is often better. |
| Podcasts and audio lessons | Turn lecture notes into a conversational script; generate an audio overview of uploaded sources; produce a study recap. |
NotebookLM
Audio Overview: ElevenLabs, Descript, Adobe Podcast (enhance speech) |
Listen once as a faculty member before assigning audio to students—tone and factual slips are easy to miss in a generated dialogue. |
| Case vignettes and problem-based learning scenarios | Generate age, setting, labs, and branching questions; ask for “red herrings” and a teaching debrief. | ChatGPT, Claude, Gemini; NotebookLM if you ground cases in a guideline PDF | Specify the year of study, available tests in a district hospital vs a tertiary centre, and local disease mix. Invented lab values must be internally consistent. |
| OSCE stations | Draft station instructions, simulated-patient scripts, checklists, and feedback phrases. | ChatGPT, Claude, Gemini | Align with your college’s OSCE format. Have two faculty members score a pilot station before using it in a real exam. |
| MCQs and item writing | Draft one-best-answer items, distractors, and rationales mapped to a learning objective. | ChatGPT, Claude, Gemini; Quizlet for student practice sets | AI writes plausible but may generate flawed items (two correct answers, cueing, outdated drugs). Human item-review is mandatory for high-stakes tests. |
| Syllabus and learning-objective mapping | Map a lecture following the curriculum, and suggested teaching methods. | ChatGPT, Claude, Gemini; NotebookLM on your existing curriculum PDF | Paste your college’s actual outcomes and the curriculum to get a mapping. |
AI can speed literature search, protocol drafting, code sketching for analysis, writing and other tasks. It cannot replace a registered protocol, a statistician, institutional review board, or YOU.
| Task / use case | What to use AI for | Tools (commonly used) | Tips & caveats |
|---|---|---|---|
| Literature search and synthesis | Clarify PICO, suggest keywords and MeSH, summarise themes, and list gaps—then you search properly. | PubMed, PMC, Google Scholar, Elicit, Consensus, Semantic Scholar, ResearchRabbit, Connected Papers, Perplexity | Keep a search log (database, date, string). Do not let a chatbot invent a “systematic review” from a handful of papers. |
| Screening for systematic reviews | Assist title/abstract screening after you set inclusion criteria; not a substitute for dual human screening on high-stakes reviews. | Rayyan, Covidence (often paid / institutional), ASReview | Document how AI was used. Follow PRISMA and your protocol; report assistance transparently. |
| Hypothesis generation and study design | Brainstorm mechanisms, confounders, sampling frames, and “what would falsify this?” questions. | ChatGPT, Claude, Gemini; Elicit for related empirical work | Treat outputs as a whiteboard. Feasibility in your hospital, sample size, and ethics come next with human experts. |
| Protocol and ethics application drafting | Turn notes into a structured protocol, consent language (plain English/Bangla draft), and a risk section to edit. | Claude, ChatGPT, Gemini; SPIRIT / EQUATOR checklists as your outline | IRB language must match local forms. Never paste identifiable records. Have a supervisor and IRB secretary review. |
| Reporting-guideline mapping | Map a draft manuscript to CONSORT, STROBE, PRISMA, CARE, or COREQ item-by-item and list missing pieces. | Claude or ChatGPT with the checklist pasted; EQUATOR Network | Download the official checklist. Models mix versions of guidelines. |
| Data analysis support | Explain a test, draft R or Python code, comment a script, or help interpret output you already generated. | ChatGPT (Advanced Data Analysis, paid), Claude, Google Colab, RStudio / R, Jupyter | De-identify datasets. Do not upload full patient data to public tools. A statistician should still review design and inference. |
| Manuscript drafting | IMRaD scaffolding, tightening Methods, language editing, cover letters, and response-to-reviewer drafts. | Claude (long documents), ChatGPT, Gemini, Paperpal, Grammarly | You remain the author. Invented references are common. Journals increasingly require AI-use disclosure. |
| Citation management | Clean metadata, convert styles, write BibTeX, find a DOI—after you have the real paper. | Zotero, Mendeley, EndNote, Scite | Never ask a chatbot to “add ten recent references.” Import from PubMed/DOI into Zotero. |
| Grant abstracts and aims | Compress a protocol into specific aims, significance, and a one-page concept note. | Claude, ChatGPT, Gemini | Funders care about feasibility and local need. Keep numbers you can defend. |
| Preprint and journal workflow | Check journal aims, draft a cover letter, and prepare a preprint checklist (conflicts, data availability). | Journal sites; medRxiv, Research Square, SSRN | Confirm co-author approval before any preprint. AI cannot know a journal’s current policy from memory alone—read the site. |
Use these tools to learn, draft, and explain—not to diagnose or prescribe without a licensed clinician reviewing the output. De-identify every data you upload.
| Task / use case | What to use AI for | Tools (commonly used) | Tips & caveats |
|---|---|---|---|
| Clinical documentation drafts (de-identified) | Turn a teaching case into a history and physical examination (H&P), progress note, or discharge-summary structure for learning documentation skills. | ChatGPT, Claude, Gemini, Microsoft Copilot | Strip names, addresses, phone numbers, dates of birth, and hospital IDs. Do not paste from the hospital patient record into a public chat. |
| Differential diagnosis as a learning exercise | Ask for a ranked differential, “what would change your mind”, and which tests discriminate, then compare with a textbook or a senior physician. | ChatGPT, Claude, Gemini | Educational framing only. Models miss rare but critical diagnoses and local disease patterns (e.g. enteric fever, dengue, TB). |
| Patient-facing education | Plain-language handouts, teach-back questions, and pictogram briefs at an appropriate literacy level. | ChatGPT, Claude, Gemini, Canva; CDC health literacy principles | Review Bangla translations carefully. Avoid alarming or false claims. Match advice to what your healthcare facility actually offers. |
| Drug information and interactions | Use AI only to phrase a question; confirm in a verified database. | Drugs.com interaction checker, DrugBank, WHO ATC/DDD, UpToDate where licensed | Do not rely on a chatbot for dosing, especially in paediatrics, pregnancy, or renal/hepatic impairment. |
| Guidelines and evidence at the point of learning | Find a guideline, then read it. Some medical-adjacent tools summarise literature. You still need to open the source. | PubMed, WHO, NICE, OpenEvidence (not available in Bangladesh), UpToDate if your institution subscribes | National and hospital protocols will always override these summaries. |
| Exam and board-style study | Flashcards, spaced-repetition decks, “explain this as if I failed it yesterday”, and mixed-question drills. | Anki, Quizlet, ChatGPT / Claude / Gemini for explanations | Generated cards often encode errors. Tag cards you have verified. Studying wrong content is worse than studying less. |
| Rounds presentations and journal club | Structure a 5-minute case, extract PICO from a paper, list strengths/limitations, and draft discussion questions. | NotebookLM on the PDF, Claude, ChatGPT, Perplexity | Read the paper yourself. Ask the model to quote page-level claims from the PDF you uploaded, then check those quotes. |
| Logbook reflections and workplace-based learning | Turn rough notes into a structured reflection (what happened, what you learned, what you will do differently) without inventing clinical facts. | ChatGPT, Claude, Gemini | Keep reflections honest and de-identified. Seniors can usually tell generic AI prose; use it to organise, not to fabricate experience. |
These workflows help teams teach and research together without turning every meeting into extra writing work.
| Task / use case | What to use AI for | Tools (commonly used) | Tips & caveats |
|---|---|---|---|
| Email, letters, and administrative writing | Tone adjustment, shorter emails, meeting invitations, and polite follow-ups. | ChatGPT, Claude, Gemini, Microsoft Copilot | Remove student or patient identifiers from examples you paste. |
| Meeting notes and action lists | Turn rough notes or a transcript into decisions, owners, and deadlines. | Descript, Otter, Gemini / ChatGPT on a transcript you control | Recording meetings needs consent. Local hospital policy may forbid cloud transcription of clinical discussions. |
| Team knowledge from PDFs and guidelines | Build a shared notebook of SOPs, lectures, and papers the model must stay within. | NotebookLM, Claude Projects, ChatGPT Projects | Control who can add files. A shared notebook with outdated guidelines will confidently teach the wrong dose. |
| Slide and handout feedback | Ask for clarity, missing learning objectives, accessibility (alt text, contrast), and timing. | Claude, ChatGPT, Gemini | Upload a PDF export rather than a huge .pptx when the tool allows files. |
Start with free web versions of Gemini, NotebookLM, ChatGPT, Claude, and Perplexity. Paid plans help heavy users (long documents, image generation, large and multi-step data analysis tasks) but are not required to learn. Local or open-weight tools (for example via Ollama) only make sense if you have a capable computer.
Do not paste patient names, addresses, phone numbers, national IDs, screenshots or pictures of medical records, or unpublished identifiable research data into public tools. De-identify teaching cases. Follow your hospital, college, and institutional data policies. If a tool’s servers are outside Bangladesh, assume you do not control where the text is stored.
Cross-check clinical and research claims with textbooks, national or WHO guidelines, and primary papers. Models are weaker on Bangladesh-specific epidemiology, drug availability, and Bangla medical language. Ask a bilingual colleague to review patient-facing Bangla translations.
Students: using AI to understand a topic is different from submitting AI text as your own assignment or
exam answer. Follow your college's rules; when in doubt, disclose.
Faculty and researchers: many
journals now expect a short statement if AI assisted writing. You remain responsible for every citation
and every clinical claim.
This page is an educational resource. It is not medical advice, not a clinical decision-support system, and not a complete catalogue of products. Named tools are examples of what people commonly use; listing them is not an endorsement, and absence is not a judgement, and I am not affiliated with any of the tools listed. Features, prices, and privacy policies may change. I shall try to keep this page updated as much as possible, but always refer to the official website for the most up-to-date information. Last updated September 2026.
If a link is broken or you want to suggest a resource for medical teachers or trainees in Bangladesh, please use the contact page.
Answer a few questions — we suggest tools from this page.
Never paste patient identifiers or photographs of patient record into public tools.