Meet the Builders: PathSetu — a Cloud Run AI mentor for first-generation students in India
Meet the Builders: 1000 Builders. 1000 Stories.
Program: Gen AI Academy APAC Edition
I am Abhishek Kumar. I am in Google Cloud Gen AI Academy APAC Cohort 3. This is my Meet the Builders story. Early-stage is allowed. A real local problem is not optional.
Watch the story
PathSetu Meet the Builders video — 1080p, about 4 minutes. Same story as this write-up. Google Drive link will be added here after upload (Anyone with the link).
The local problem
In India, career advice still concentrates in a few metro campuses and a few LinkedIn circles. First-generation students in tier-2 and tier-3 cities — Indore, Ranchi, Coimbatore, Guwahati, Nashik — often have the degree, not the map.
They get stuck on the same three questions:
- What role am I actually hiring for — PM, BA, operations, data — and what does “good” look like in year one?
- How do I turn college projects into a portfolio a hiring manager will trust?
- What should I practice this week, not “learn product management” as a slogan?
Generic chatbots hallucinate salary numbers, invent company processes, and give advice that does not match campus placements, family constraints, or Hindi/English mixed interviews.
That is a local access problem, not a content shortage.
What I am building
PathSetu is a student-facing AI mentor on Cloud Run. It answers in plain language, grounded in a curated career knowledge base (role definitions, portfolio rubrics, interview drills, weekly plans) — not the open internet.
If the question needs a human (a specific college placement cell, mental health, a scholarship office), it says so and stops.
Mapped to the Cohort 3 labs:
| Cohort 3 lab | PathSetu equivalent |
|---|---|
| Personalized customer experience (ADK + RAG) | Mentor agent that retrieves role maps, portfolio checklists, and interview drills from approved notes |
| Location intelligence (Gemini + BigQuery / MCP) | Where to run campus workshops first: city/college clusters with high demand and weak mentor supply |
| Operations productivity agent on Cloud Run | Placement-cell / campus-club assistant for session plans, follow-ups, and weekly practice queues |
Architecture (production path)
Path: Student → ADK + Gemini → RAG knowledge pack → Cloud Run → grounded answer, or human handoff.
| Step | What happens |
|---|---|
| 1. Ask | Student writes in plain language |
| 2. Reason | ADK + Gemini on Cloud Run |
| 3. Ground | Retrieve from the curated pack — not the open web |
| 4. Decide | Cite and answer, or stop and hand off to a counsellor / college office |
| 5. Locate (Track 2) | BigQuery + MCP: demand vs mentor supply by city/college |
| 6. Ops (Track 3) | Placement-cell agent: session plans and weekly queues |
Rules I will not break:
- No invented salaries, headcount, or “guaranteed placement” claims
- Cite the retrieved note when the answer is process-specific
- Escalate when the user needs a human counsellor or official college office
- Keep student PII out of prompts; treat chat logs as sensitive
Why this matters in APAC
Mentor access is uneven. A grounded agent on Cloud Run will not replace a good teacher. It can give a first-generation student a repeatable weekly plan at 11pm, in the language they actually use, without waiting for a metro mentor to reply.
Comments
Post a Comment