8  Guide 5 — AI PhD Supervisor Agent

8.1 “Virtual Research Mentor”

8.2 The problem

Many PhD students struggle because:

  • supervisors are overloaded and unavailable (one meeting a month is common),
  • research direction drifts without anyone noticing,
  • there is no continuous, structured feedback,
  • students lack methodology and writing coaching,
  • the defence arrives with little rehearsal.

Completion rates in many programmes are below 50%, and average time-to-degree keeps growing.

8.3 The solution

An AI mentor (assistant to the supervisor, coach to the student) that keeps long-running memory of the student’s project:

Agent Role Output
Research Planning Agent Creates a 3-year PhD roadmap with milestones, deliverables, risks roadmap.md / roadmap.json
Weekly Progress Agent Student submits a weekly update; agent scores progress vs. plan, flags problems, sets next-week goals progress_log.jsonl, weekly report
Critical Reviewer Agent Challenges assumptions, methodology and novelty of the current work critique.md
Presentation Agent Generates conference slides outline, poster text and a mock-defence Q&A slides.md, poster.md, defence_qa.md

8.3.1 Example

Week 12 update: “Finished the RGB image pipeline; model accuracy 87%; hyperspectral data delayed.”

Agent evaluation:

  • Progress: 80% of the milestone. On track: Aim 1. At risk: Aim 2 (data delay).
  • Problems: “Need more validation experiments — single split, no cross-validation.”
  • Next week: (1) 5-fold CV, (2) contact core facility for hyperspectral timeline, (3) draft methods section.

8.4 Architecture — memory is the product

student profile + topic ─► Research Planning Agent ─► roadmap (milestones)
                                                          │ stored in memory/
weekly_update.md ─► Weekly Progress Agent (reads roadmap + last N updates) ─► evaluation (structured)
                                                          │ appended to progress_log.jsonl
       "critique" ─► Critical Reviewer Agent (reads everything) ─► critique.md
    "presentation" ─► Presentation Agent ─► slides / poster / defence Q&A

Memory design

  • A project folder per student (memory/<student_slug>/) with profile.json, roadmap.json, progress_log.jsonl.
  • Each call receives a compact context summary (roadmap + last 4 updates), not the whole history → cheap and focused.
  • Nothing is deleted; the log is the audit trail the supervisor can read.

8.5 Build steps

Step File What you learn
1 memory.py Simple file-based memory (JSON/JSONL) with load/append/summarise
2 agents.py Roadmap (structured), weekly evaluation (structured), critic, presentation
3 main.py Sub-commands: init, weekly, critique, present

Run:

python agent_05_phd_supervisor\main.py init --student "Ali" --topic "Multimodal AI for drought resilience in wheat" --years 3
python agent_05_phd_supervisor\main.py weekly --student "Ali" --update "Finished RGB pipeline, accuracy 87%, hyperspectral data delayed"
python agent_05_phd_supervisor\main.py critique --student "Ali"
python agent_05_phd_supervisor\main.py present --student "Ali" --type defence

8.6 Product extensions

  • Supervisor dashboard: all students, traffic-light status, one-click summaries before meetings.
  • Calendar & reminders (WhatsApp/Email/Slack) for weekly check-ins.
  • Document review: upload a chapter → structured feedback against the roadmap.
  • Mock viva with voice; question bank per discipline.
  • Wellbeing signals: detect stalled progress and suggest supervisor intervention (with consent).
  • Institutional analytics: cohort completion risk for graduate schools.

8.7 Market value

Customer Value Price
PhD students Weekly structured feedback, faster completion $10–20/month
Supervisors Prepared students, less admin included in lab plan ($2k–5k/year)
Graduate schools Higher completion rates, accreditation evidence $30k–100k/year
Online/remote PhD programmes Scalable supervision support custom

The stickiest agent: once a student’s 3-year memory lives here, they do not leave.

8.8 How to sell it

  1. Graduate-school pilots: 20 students for one semester; report completion-risk reduction.
  2. Supervisor champions: give supervisors a free dashboard; students pay or the department pays.
  3. HEC / accreditation angle (Pakistan and similar): supervision logs are required evidence — the agent produces them automatically.
  4. Bundle with the other four agents as the “PhD success plan”.
  5. Community: PhD-student WhatsApp/YouTube communities (Codanics-style) as the distribution channel.

8.9 Risks & mitigations

Risk Mitigation
Perceived as replacing supervisors Positioning + supervisor dashboard; supervisor stays in the loop
Over-reliance / wrong advice Every output carries a “verify with your supervisor” note; critic agent challenges its own advice
Student data privacy Local/on-prem memory, encryption, opt-in sharing with supervisor
Generic coaching Memory-grounded prompts; discipline-specific templates

8.10 90-day roadmap

  • Weeks 1–2: CLI with file memory (this repo) → Weeks 3–6: web app + supervisor dashboard → Weeks 7–10: chapter review + reminders → Weeks 11–13: mock viva & cohort analytics.

Codanics — Python ka Chilla: Build AI Agents

  • Course: https://codanics.com/courses/python-ka-chilla-build-ai-agents/
  • Website: https://codanics.com · YouTube: @codanics
  • WhatsApp channel: https://whatsapp.com/channel/0029Va7nRDq3QxRzGqaQvS3r
  • Complete free course: https://www.youtube.com/watch?v=UE3o-7KRNXs&list=PLaQmdklCcX40