Goal & SIP navigator
About this agent
Goal & SIP Navigator is a rebalancing and goal-tracking co-pilot for mutual fund advisors, MFDs and RIAs, aimed at replacing reactive market-timing with a systematic capital-preservation process. A pure-Python rules engine — not the LLM — decides what's actually true: whether a client's equity/debt split has drifted past a threshold from their target allocation, whether the Nifty PE ratio has crossed an overvaluation guardrail, and whether any of a client's goals has entered its glide-path window, the months before its target date when equity exposure should start shifting into lower-volatility debt. Only once those checks run does the agent call the LLM, and only to narrate the findings for the advisor — the rule outputs are passed into the prompt explicitly so the model is never left to infer the numbers itself. It also recommends an annual SIP step-up sized to the lesser of income growth and inflation-plus-a-margin, and projects whether that step-up actually closes the gap to a goal's target corpus. A separate fee audit quantifies the commission drag of staying in a Regular Plan instead of a Direct Plan, for the client-facing transparency conversation. Session memory holds one review's temporary simulations and chat feedback; once an advisor approves a rebalance, it's logged atomically into a per-firm CLIENT_GOALS project and the session closes, promoting it to long-term history. An optional daily scheduler can sweep an entire book of clients and raise an alert for every one that breaches.
What changed with Zenmem?
The same agent, built twice against the same contract — once on Zenmem, once on MongoDB + LangChain/LangGraph.
Before → after
Before With Zenmem
What the team gained
- Live valuations from the connected database and the client's long-term goal history arrive in the same model call, so a review needs no staging step between the two.
- An approved action is committed into the client's goal scope inside a transaction, so goal timeline, thresholds and allocation log move together.
- A rebalance simulation stays in session scope, so modelling an action a client declines leaves the record untouched.
- IPS thresholds live beside the goals they guard, rather than in a config file that drifts from the book.
- A new goal type is a document, not a column added across every client.
How memory is scoped
Session scope holds one goal-safeguard review — a temporary rebalance simulation and any investor chat feedback for that sitting. connectedDB pulls live valuations and passMemory pulls this client's long-term goal history into the same callLLM call. Once an advisor approves an action, it's written into project scope, projectId CLIENT_GOALS, inside a transaction — goal timelines, IPS thresholds and allocation logs live there long-term — and the session is then closed.
How it works
Rules decide first; the LLM narrates after.
Run the rules engine
Pure Python checks: drift from target allocation, a Nifty PE valuation guardrail, and each goal's glide-path window.
Suggest a safeguard shift
If any rule breaches, a fixed, auditable slice of equity is shaved into debt — capped so equity never drops below 30% from this rule alone.
Narrate for the advisor
callLLM turns the rule findings into a Capital Preservation Alert, with connectedDB and passMemory bringing in live valuations and goal history.
Log the action atomically
An approved rebalance is written into project memory inside a client.transaction(), then the session closes and promotes it to long-term storage.
Sweep the book (optional)
A daily scheduler re-runs the safeguard check across every client and raises an alert for each breach.
What it does
The rules engine and fee audit behind the LLM narration.
Capabilities
- Flags portfolios when equity/debt allocation drifts beyond a configurable threshold (default 10%) from the client's target allocation.
- Applies an objective Nifty PE valuation guardrail (default 22) to recommend tactical de-risking, rather than emotional market-timing.
- Tracks each goal's glide-path window — once a goal is within a configurable number of months (default 24) of its target date, it's flagged for a shift into lower-volatility instruments.
- Recommends an annual SIP step-up as the lesser of income growth and inflation-plus-margin, then projects whether the stepped-up SIP actually closes the gap to the goal's target amount.
- Audits the commission drag of a Regular Plan against a Direct Plan over the holding period, for the client-facing transparency conversation.
- Logs an approved rebalance into project memory inside a client.transaction(), so a partial write never lands if something fails mid-way.
- Runs an optional daily sweep across a whole book of clients, raising an alert for every client whose safeguard check breaches.