I’m Toka Nani, an Applied AI and Cloud Engineer who takes products from idea to deployment—not just to a notebook. My work spans LLM cost optimization, responsible AI, and multilingual decision systems, using Python, FastAPI, React, Docker, and Google Cloud.
Google Gemini Student Ambassador · AWS Certified Solutions Architect – Associate · B.Tech CSE ’28
| Applied AI engineering | Cloud & platform engineering | Product engineering |
|---|---|---|
| Gemini, Vertex AI, embeddings, fairness metrics, and evaluation-driven LLM workflows. | Containerized services, Cloud Run, Firestore, Cloud Storage, and CI/CD with GitHub Actions. | React and TypeScript interfaces backed by tested Python APIs and practical data pipelines. |
Currently seeking: Software Engineering, Applied AI, Cloud, and DevOps internships. Open to remote work and relocation. Based in India · UTC+5:30.
The problem: Repeated prompt context increases token usage and makes LLM applications more expensive to operate.
What I built: Middleware that combines Gemini text-embedding-004 vectors with exponential recency-decay compression before each completion call.
Measured result: ~40% fewer prompt tokens at ~96% answer quality — published benchmark in the repo (benchmarks/RESULTS.md: 370 → 221 tokens), reproducible with one command.
FastAPI Gemini 1.5 Flash Embeddings React 18 Render
Live product · API documentation · Source code
The problem: Hiring and credit-scoring systems can produce inequitable outcomes without a clear, reviewable audit trail.
What I built: A forensic auditing pipeline that streams decision logs through demographic-parity and equalized-odds checks, then uses Gemini to generate compliance reports mapped to UN SDG-5 and SDG-10.
Measured result: Compliance reports generated in under 30 seconds. Designed and built solo for Google Solution Challenge 2026.
Vertex AI Gemini Flask Firestore Cloud Run Docker
Decision logs → Cloud Storage → Flask on Cloud Run → Fairness evaluation → Gemini report → Firestore audit trail
What I built: A multilingual civic decision-intelligence platform. Gemini detects language and categorizes complaints; a deterministic four-factor scoring engine ranks demand, infrastructure gaps, vulnerability, and investment overlap.
Engineering proof: 48 passing pytest tests across the CivicPulse backend.
FastAPI React TypeScript Vite Gemini
What I built: A nonpartisan voter-readiness assistant combining a deterministic decision-tree engine with Gemini-powered Q&A, containerized and deployed on Cloud Run for PromptWars.
React TypeScript Gemini Docker Cloud Run
More engineering work
- Mutual Fund Analytics Platform — Bluestock capstone: Nine-table SQLite star schema with 86,000+ rows; CAGR, Sharpe, Sortino, Alpha-Beta, VaR, and CVaR analytics in a four-page Streamlit dashboard. Source code
- JanVoice AI: Role-based grievance portals for citizens, MPs, and administrators, with Gemini-generated daily constituency briefings. Live product
| Timeline | Role or milestone | What it demonstrates |
|---|---|---|
| May 2026 — Present | Google Gemini Student Ambassador, Google India | Leading AI workshops and developer sessions; communicating technical concepts to campus communities. |
| Sep — Oct 2026 | Data Analyst Intern, Bluestock Fintech | Built an end-to-end mutual-fund analytics pipeline with live AMFI NAV integration. |
| 2026 | Google Solution Challenge | Designed and shipped BiasGuard AI as a solo builder. |
| 2026 | PromptWars Virtual, Challenge 3 | Placed in the Top 400. |
| 2026 | Anvil @ Ascent, Scaler School of Technology | Grand Finale qualifier. |
Professional simulations: JPMorgan Software Engineering · Walmart Global Tech Advanced SWE · Tata iQ GenAI Analytics.
Certifications & credentials
| Certification | Issuer | Verification |
|---|---|---|
| AWS Solutions Architect – Associate | AWS | Verify credential |
| Google AI Essentials | Google / Coursera | Verify credential |
| Claude Code in Action | Anthropic | Verify credential |
| Google Cloud Gen AI Academy APAC, Cohort 3 | Google Cloud × Hack2skill | Verify credential |
| Amrita Agentic Leap 2026 | Amrita Vishwa Vidyapeetham | Verify credential |
| Build with AI Bootcamp, Chennai | Google for Developers | Verify credential |
| Google Solution Challenge 2026 | Google × Hack2skill | Verify credential |
| PromptWars Virtual, Challenge 1 | Google for Developers | Verify credential |
| Capability | Tools |
|---|---|
| AI & ML | Gemini API · Vertex AI · embeddings · prompt engineering · fairness metrics |
| Languages | Python · TypeScript · JavaScript · SQL |
| Backend & data | FastAPI · Flask · Node.js · Pandas · SQLAlchemy · SQLite · Streamlit |
| Frontend | React · Vite · Tailwind CSS |
| Cloud & DevOps | Google Cloud Run · Cloud Storage · Firestore · Docker · Render · GitHub Actions · Git |
- Start with the decision or user outcome, not the model.
- Keep critical logic deterministic and tested; use AI where it adds clear value.
- Deploy early and instrument the result—cost, latency, quality, or fairness.
- Make the system understandable through documentation, audit trails, and usable interfaces.
