01
AI Systems & Retrieval
LLM pipelines, hypothesis-driven retrieval, source evaluation, grounding, provenance, and evidence quality.
Diramu Godana
Computer Science & Economics at Harvard
Software engineer building AI systems, data pipelines, and intelligent products across labor-market intelligence, recommender systems, education, and healthcare.
Get in TouchWork Experience
Software Engineering Intern
May 2026 — Aug 2026- Built the Context Agent for an AI labor-market intelligence platform, converting anomalies surfaced from ~42M member profiles and 7.6M job postings into structured investigations and grounded evidence for generated labor-market articles.
- Engineered source scoring, deduplication, support/challenge balancing, provenance tracking, and bundle-sufficiency checks, increasing hypothesis/query plans meeting the highest quality tier from 5.7% to 86.4% and usable evidence bundles from 2.1% to 90%.
- Built an end-to-end Economic Graph pipeline that identified top-growth industries, ranked their fastest-growing companies across size bands, validated each signal with postings, views, applications, and role mix, and generated evidence-backed articles; surfaced 111 Tier-1 opportunities from 407 candidates across 9 industries.
- Extended LinkedIn’s HERO Generative Recommender evaluation stack across PySpark, Flyte, and OpenHouse, adding an underserved-job benchmark, slice-aware semantic metrics, and leakage controls for 1.08M held-out member-job pairs; evaluated 3,982 jobs and found a 17.3 percentage-point Good-Fit P@200 gap between underserved jobs and the existing benchmark.
Harvard College GenAI Research Program
Software Engineering Intern
Jun 2025 — Aug 2025- Engineered a bilingual RAG tutor translating English textbook material into Swahili, improving comprehension by ~20% for students in rural Kenya.
- Increased retrieval accuracy from 70% to 95% by processing 1K+ textbook chunks with LangChain, Pinecone, and GPT-4 Turbo.
- Optimized PostgreSQL schemas and RESTful FastAPI endpoints, reducing query latency from 5s to under 2s.
- Built summarization, revision Q&A, and free-form tutoring modes using a Next.js, Tailwind, and FastAPI stack.
- Piloted the platform with 56 students, providing dual-language output aligned with classroom curriculum.
Kerugoya County Referral Hospital — Kerugoya, Kenya
Software Engineering Intern / Priscilla Chan Fellow
Jun 2024 — Aug 2024- Built a hospital management portal automating records, billing, and scheduling for 1,000+ patients, cutting wait times by 30%.
- Developed ERP/EMR modules in Java and Python with Frappe, deployed via Docker, and integrated MySQL and REST APIs.
- Worked across feature design, debugging, testing, and deployment, reducing post-deployment defects by 25%.
Harvard Undergraduate Machine Intelligence Community (HUMIC)
Project Lead & Web Scraping Engineer
Sep 2024 — Present- Promoted from Fellow to Project Lead while developing a multi-agent self-learning assistant using CrewAI.
- Built Python web-scraping pipelines with BeautifulSoup, Requests, and BraveSearch API to collect and structure educational content.
- Coordinate profiling, ranking, and roadmap work across teams building personalized learning systems.
Harvard WECode
Associate Director of Engineering
Sep 2025 — Present- Co-developed WECode’s engineering platform supporting the world’s largest student-run women-in-tech conference with 1,200+ attendees.
- Built authentication, registration, mentorship, and event workflows using React and Supabase.
- Lead technical planning, code reviews, and deployment work across the engineering team.
Featured Projects
Adaptive Swahili-English AI Curriculum Tutor
Bilingual AI tutor for Form 1 Biology built with LangChain, Pinecone, GPT-4 Turbo, FastAPI, and Next.js. Supports curriculum-grounded summarization, revision Q&A, and free-form tutoring in both English and Swahili.
WECode Conference Platform
Full-stack conference platform supporting authentication, registration, mentorship matching, and event workflows for Harvard WECode. Built with React, Supabase, and modular backend services.
AI-Powered Multilingual Flashcard Generator
Interactive learning application that generates English and Swahili flashcards, AI study summaries, and text-to-speech audio, with caching and session-state optimizations for faster responses.
Nairobi After — Immersive Web Narrative
Interactive digital narrative built with HTML, CSS, and JavaScript, combining scroll-based storytelling, glitch animations, ambient soundscapes, and dynamic chapter transitions.
About Me
I'm a Harvard student studying Computer Science and Economics, interested in building AI systems that turn complex data into useful, grounded products. Most recently, I interned at LinkedIn, where I worked across labor-market intelligence, Economic Graph data pipelines, and recommender-system evaluation. I built systems that transformed large-scale labor-market signals into grounded evidence for generated insights, developed an end-to-end pipeline for identifying high-growth company opportunities, and extended evaluation infrastructure for LinkedIn's HERO Generative Recommender. My earlier work includes bilingual AI tutoring systems, healthcare software, and multi-agent learning tools. I also serve as Associate Director of Engineering for Harvard WECode and as a Project Lead within the Harvard Undergraduate Machine Intelligence Community.
My Areas of Specialization
02
Data & ML Infrastructure
Python, SQL, PySpark, Flyte, OpenHouse, large-scale data pipelines, and reproducible evaluation workflows.
03
Recommender Systems Evaluation
Benchmark construction, slice-aware semantic metrics, train/eval separation, and large-scale model validation.
04
Backend Engineering
FastAPI, PostgreSQL, MySQL, REST APIs, Docker, and production-oriented service design.
05
Full-Stack Development
React, Next.js, Tailwind, Supabase, and end-to-end development of user-facing applications.
06
Applied AI
AI systems spanning labor-market intelligence, education, healthcare, and personalized learning.