Research Communication
Jul 28–31 · Final poster + presentation · July 31: REU Closing Ceremony
Week 6 Agenda
Presentation Tips
The 90-Second Pitch
When a visitor approaches your poster, you have 90 seconds before they move on. Practice this: "I studied X. The question I investigated was Y. I found Z. The implication is W." That's 4 sentences. Rehearse until it's natural.
What Visitors Ask
- "How does ComplianceGPT work?" → Give the 3-sentence pipeline description
- "Why does accuracy matter?" → Healthcare compliance errors cost patients and hospitals
- "Did you expect this result?" → Be honest; surprises make better talks
- "What would you do next?" → Have one concrete next step ready
Figures and Tables
Every figure needs:
- A clear axis label with units
- A caption that states the main takeaway
- High enough contrast to read from 6 feet away
- No unnecessary decoration (chartjunk)
Caption formula:
"Figure 1: Few-shot prompting increases extraction accuracy. [Description of what's shown.] Few-shot prompting achieves 82% accuracy vs. 74% for zero-shot across 137 HIPAA scenarios."
Lead with the takeaway, not the description.
# Minimal matplotlib for a clean bar chart
import matplotlib.pyplot as plt
methods = ["Zero-shot", "Few-shot", "CoT"]
accuracy = [0.74, 0.82, 0.79]
fig, ax = plt.subplots(figsize=(6, 4))
bars = ax.bar(methods, accuracy, color=["#990000", "#003366", "#f0a500"])
ax.set_ylim(0, 1.0)
ax.set_ylabel("Accuracy")
ax.set_title("Extraction Accuracy by Prompting Strategy (n=137)")
for bar, val in zip(bars, accuracy):
ax.text(bar.get_x() + bar.get_width()/2, val + 0.01,
f"{val:.0%}", ha="center", fontsize=10, fontweight="bold")
plt.tight_layout()
plt.savefig("figures/accuracy_by_strategy.pdf", dpi=150)
What Comes Next
Letter of Recommendation
If you want a letter from me, send me a request email within 2 weeks of the program ending. Include: where you're applying (grad school? internships?), deadline, and what you'd like me to emphasize. I'll write a specific, substantive letter for anyone who engaged seriously with their project.
Paper Submission
If your project has a strong result, I will invite you to co-author a paper. I handle submission logistics. Your job: clean up the write-up and respond to my revision requests. Target venues: ACL, EMNLP, NAACL student research workshop, or an AI + Law workshop.
Graduate School
REU experience is strong evidence for PhD programs. When you apply, you can say: "I ran experiments comparing LLM prompting strategies on 137 real HIPAA compliance scenarios at Stony Brook University." That's a concrete research contribution — more valuable than coursework.
Stay in Touch
Send me a LinkedIn connection before you leave. Let me know where you end up. If you're applying to PhD programs in NLP, AI safety, or trustworthy AI in the next 1–3 years — reach out. I'll be happy to advise.
Final Checklist Before July 31
- Poster submitted to REU coordinator (print-ready PDF, 48×36 inches)
- Final write-up submitted to me (2+ pages)
- Results CSV committed to GitHub:
results/[yourname]_final.csv - 90-second pitch rehearsed and timed
- Experiment notebook clean and runnable from top to bottom
- OPTIONAL: draft abstract for a potential workshop paper submission
Thank you for being part of this lab this summer.
Research is slow, uncertain, and rewarding. The fact that you finished a project, ran real experiments, and can explain your findings to a room full of strangers — that's the whole point.